What changed, and when

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: the FAQ's list of successor models now reads Claude Opus 5.5, GPT-6 Astra, Gemini 4 Argon and the Gemini 3.8 Flash line, since Gemini 3.1 Pro has also been superseded. The Gemini 3.1 Pro AA-Omniscience row keeps the measured model and is now dated May 2026; no benchmark figure was renamed.

Referenced-model links updated to the October 2026 lineup: GPT-5.6 Terra to GPT-6.1 Terra, Claude Opus 5 to Claude Opus 5.5. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: GPT cost row moved to GPT-6.1 Terra ($1/$5) and recalculated from ~$3 to ~$1.30 per million chars (same token volume, 6/14 of the old per-token cost); Claude cost row renamed to Sonnet 5.5 at an unchanged ~$3.50. Matrix quality rows keep the September 2026 tested versions; added an update notice saying so.

Referenced-model links updated to the October 2026 lineup: Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.6 Terra to GPT-6.1 Terra. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: the Claude Team model-under-the-hood cell now reads Sonnet 5.5 and Opus 5.5 instead of Sonnet 5 and Opus 5. Seat prices are plan prices, not per-token, so nothing was recalculated; the 'Opus needs a paid plan' note is carried over from the September check, not re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: inference-cost row now names Claude Sonnet 5.5 and GPT-6.1 Terra. Range kept at ~$0.001–$0.005 per ticket because Sonnet 5.5 holds Sonnet 5's $2/$10; GPT-6.1 Terra ($1/$5, down from GPT-5.6 Terra's $2/$12) sits at the low end, which the row now says. Hybrid and monthly totals re-checked against that range and unchanged.

Referenced-model links updated to the October 2026 lineup: Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.6 Terra to GPT-6.1 Terra. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: Claude flagship and 1M-context models now named as Opus 5.5 / Sonnet 5.5 (plus Fable 5.1); OpenAI context row now cites the GPT-6 API models only, as GPT-5.6 has been replaced. No prices on this page depend on API model rates, so none were recalculated.

Referenced-model links updated to the October 2026 lineup: Claude Opus 5 to Claude Opus 5.5. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: long-context prerequisite now names Claude 5.5 and GPT-6.x. Per-contract cost recalculated for a 10–25k-token contract with ~2–3k output tokens: $0.002–$0.015 on Claude Haiku 5.5 / GPT-6 Luna / Gemini 3.5 Flash-Lite (Haiku 5.5 counted at +30% for its tokenizer), $0.02–$0.08 on Claude Sonnet 5.5 / GPT-6.1 Terra; was $0.05–$0.30 with no model named. Added a note that Haiku 5.5 bills 5x above 100k prompt tokens, which a ~77k-token contract already crosses on its tokenizer.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links: Qwen2.5-Coder 32B to Qwen3-Coder-Next, and added Devstral Small 2 — the two current open coding models that run on one machine. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: added an update notice naming the successor models (Claude Opus 5.5 / Sonnet 5.5, GPT-6.1 Sol, Grok 4.7, Gemini 4 Argon). The model-availability rows deliberately keep the September 2026 vendor-docs snapshot, because which tool offers or defaults to the new models has not been re-verified. No prices changed.

Referenced-model links updated to the October 2026 lineup: Claude Opus 5 to Claude Opus 5.5, Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.6 Sol to GPT-6.1 Sol, Grok 4.6 to Grok 4.7. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: recommended defaults and cost row moved to Claude Sonnet 5.5 and GPT-6.1 Terra. Per-draft cost recalculated at a 6–8k-token context pack with ~300–600 output tokens: Sonnet 5.5 stays at $0.015–$0.022 (same $2/$10 price), GPT-6.1 Terra is about half that at $0.0075–$0.011 ($1/$5, down from GPT-5.6 Terra's $2/$12).

Referenced-model links updated to the October 2026 lineup: Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.6 Terra to GPT-6.1 Terra. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: the FAQ's vision-LLM examples now name Claude Sonnet 5.5 / Opus 5.5 and Gemini 4 Argon instead of Sonnet 5 / Opus 5 and Gemini 3.1 Pro. No per-page price depended on those models, so none was recalculated.

Prose refreshed to the October 2026 lineup: LLM-fallback row now names Claude Haiku 5.5 / GPT-6 Luna / Gemini 3.5 Flash-Lite. Fallback cost recalculated at ~1,500–4,000 input + ~100 output tokens (Haiku 5.5 counted at +30% for its tokenizer): $0.0002–$0.0015 per document, down from $0.001–$0.005. Knock-on totals: 10,000-document backfill $1–8 (was $5–25), 200 docs/day $0.50–5 per month (was $3–15), header cost line to match. Embedding-cost row and flagship-only baseline unchanged.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna, GPT-4.1 to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: cost row now names Claude Haiku 5.5 / GPT-6 Luna / Gemini 3.5 Flash-Lite. Per-email cost recalculated assuming ~800–1,500 input and ~150–300 output tokens (Haiku 5.5 counted at +30% for its tokenizer): about $0.0002–$0.0012 per email, $0.03–$0.24 per day at 200 emails, about $1–7 per month. The old $10–25/month did not follow from its own per-day cost.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna, GPT-4.1 to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: flagship cost row now names Claude Sonnet 5.5 / GPT-6.1 Terra / Gemini 4 Argon, and the structured-outputs line reads 'and later' so it covers the current generation. Per-page range kept at $0.005–$0.05: Sonnet 5.5 is unchanged at $2/$10, GPT-6.1 Terra is cheaper than GPT-5.6 Terra ($1/$5 vs $2/$12), and Gemini 4 Argon's $2/$10 doubles to $4/$20 on 2026-12-01, which the row now notes. Added a referenced-model link for Gemini 4 Argon.

Referenced-model links updated to the October 2026 lineup: GPT-5.6 Terra to GPT-6.1 Terra, Claude Opus 5 to Claude Opus 5.5, Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.5 to GPT-6.1 Sol. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.6 Terra to GPT-6.1 Terra. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: vision cost row and API prerequisite now name Claude Sonnet 5.5 / GPT-6.1 Terra / Gemini 4 Argon (tools list GPT-5.5 vision to GPT-6.1 vision). Per-image range kept at $0.0005–$0.005: Sonnet 5.5 is unchanged at $2/$10, GPT-6.1 Terra is cheaper than GPT-5.6 Terra ($1/$5 vs $2/$12), and Gemini 4 Argon's $2/$10 doubles to $4/$20 on 2026-12-01, which the row now notes. Added a referenced-model link for Gemini 4 Argon.

Referenced-model links updated to the October 2026 lineup: GPT-5.6 Terra to GPT-6.1 Terra, Claude Opus 5 to Claude Opus 5.5, Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.5 to GPT-6.1 Sol. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: API price and context rows now show GPT-6.1 Sol ($2/$10) and Terra ($1/$5), Claude Opus 5.5 ($4/$20) and Sonnet 5.5, Gemini 4 Argon ($2/$10 introductory, $4/$20 from 1 December 2026); API picks moved to Sonnet 5.5, GPT-6.1 Terra and GPT-6 Luna ($0.10/$0.50); volatility notes updated for Gemini 4 Argon and the GPT-6 price reset. Update notice refreshed. Quality rankings unchanged: still the Q1 2026 evaluation of GPT-5.4, Sonnet 4.6 and Gemini 3.1 Pro.

Referenced-model links updated to the October 2026 lineup: GPT-5.6 Terra to GPT-6.1 Terra, Claude Opus 5 to Claude Opus 5.5, Gemini 3.1 Pro Preview to Gemini 4 Argon, Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.6 Luna to GPT-6 Luna, GPT-5.6 Sol to GPT-6.1 Sol. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: proprietary-frontier examples now Claude Opus 5.5, GPT-6 Astra and Gemini 4 Argon (matrix and hybrid-pattern list); the multi-GPU-server open-weight example is now DeepSeek V4.1 Flash (~763B total, up from V4 Flash's 284B), and V4 Pro is no longer called the very top of the open-weight range, since DeepSeek reports V4.1 Flash ahead of it. No break-even figure depended on a named model price, so none was recalculated.

Referenced-model links updated to the October 2026 lineup: GPT-5.6 Terra to GPT-6.1 Terra, Claude Opus 5 to Claude Opus 5.5, DeepSeek V4 Flash to DeepSeek V4.1 Flash. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Opus 5 to Claude Opus 5.5, GPT-5.6 Terra to GPT-6.1 Terra. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Sonnet 5 to Claude Sonnet 5.5, GPT-4o mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: flagship price rows now name GPT-6 Astra, Claude Opus 5.5 and Gemini 4 Argon (input still $2–$10, output now $10–$50, with Argon's 2026-12-01 rise to $4/$20 noted); small-tier row now GPT-6 Luna, Claude Haiku 5.5 and Gemini 3.5 Flash-Lite at $0.10–$0.30 input and $0.50–$2.50 output.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, Claude Opus 5 to Claude Opus 5.5, GPT-5.6 Luna to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: the open-weights quality-gap row keeps Claude Opus 5, the model it was assessed against, and now dates that comparison (September 2026) rather than implying a re-test against Opus 5.5. No prices on this page depend on the changed models.

Referenced-model links updated to the October 2026 lineup: Claude Opus 5 to Claude Opus 5.5. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: cost row moved to Claude Sonnet 5.5. Range unchanged at ~$0.01–$0.035 per draft, since Sonnet 5.5 keeps Sonnet 5's $2/$10 price (the catalog lists no tokenizer change).

Referenced-model links updated to the October 2026 lineup: Claude Sonnet 5 to Claude Sonnet 5.5. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Opus 5 to Claude Opus 5.5. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: context-window prerequisite now names Claude 5.5 and GPT-6.x. Summary costs recalculated at Claude Haiku 5.5 / GPT-6 Luna / Gemini 3.5 Flash-Lite pricing, assuming ~6–12k input tokens for a 200-message day and ~30–60k for a 1,000-message week, output at ~5–8% of that (Haiku 5.5 counted at +30% for its tokenizer): $0.001–$0.006 per daily summary (was $0.01–$0.05), $0.004–$0.03 per weekly (was $0.05–$0.25). Header API cost restated as $1–2/month for ~10 channels daily (was $5–15 at unstated 'moderate volume').

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: cost row moved to Claude Sonnet 5.5 (same $2/$10 per MTok as Sonnet 5, so the ~$0.03 per summary at 12k in / 800 out is unchanged); context prerequisite now names Opus 5.5, Sonnet 5.5 and GPT-6.x.

Referenced-model links updated to the October 2026 lineup: Claude Sonnet 5 to Claude Sonnet 5.5. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: price table now lists GPT-6.1 Terra ($1/$5), Claude Sonnet 5.5, GPT-6.1 Sol, Gemini 4 Argon ($2/$10 introductory, $4/$20 from 1 Dec) and Claude Opus 5.5 ($4/$20); context-window and long-prompt surcharge notes updated. The three worked examples were still on the older $3/$15 mid-tier rate and are recalculated at $2/$10 (support assistant ~$240 to ~$165/month, knowledge-base assistant ~$2,700 to ~$1,800/month, PDF pipeline $7,500 to $5,000/month); RAG FAQ input cost $3 to $2 per 1M-token call.

Referenced-model links updated to the October 2026 lineup: GPT-5.6 Terra to GPT-6.1 Terra, Claude Opus 5 to Claude Opus 5.5, Gemini 3.1 Pro Preview to Gemini 4 Argon, Claude Sonnet 5 to Claude Sonnet 5.5, GPT-5.5 to GPT-6.1 Sol, GPT-5.6 Sol to GPT-6.1 Sol. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: cost row now names Claude Haiku 5.5 / GPT-6 Luna / Gemini 3.5 Flash-Lite. Per-email cost recalculated at ~600 input + ~60 output tokens (Haiku 5.5 counted at +30% for its tokenizer): about $0.0001–$0.00035 per email, $0.02–$0.07 per day at 200 messages. The old $5–15/month figure did not follow from its own per-day cost; corrected to about $1–2/month.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna, GPT-4.1 to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Referenced-model links updated to the October 2026 lineup: Claude Haiku 4.5 to Claude Haiku 5.5, GPT-4.1 mini to GPT-6 Luna. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: the autonomous-agent capability example now names Claude Sonnet 5.5 / Opus 5.5 instead of Sonnet 5 / Opus 5, and Opus 5.5 is added to the referenced models. No cost figure depended on a named model.

Referenced-model links updated to the October 2026 lineup: Claude Sonnet 5 to Claude Sonnet 5.5. Page content not otherwise re-verified.

Prose refreshed to the October 2026 lineup: the largest-context-window row now names GPT-6 models (GPT-6.1 Sol, 1.05M tokens) instead of GPT-5.5. The API cost ranges are generic and still cover the current mid-tier prices, so they were not changed.

Referenced-model links updated to the October 2026 lineup: GPT-5.6 Terra to GPT-6.1 Terra, Claude Opus 5 to Claude Opus 5.5, GPT-5.5 to GPT-6.1 Sol. Page content not otherwise re-verified.

Added hardware requirements (Mac support, memory, NVIDIA). Source: the model card and README (Mac figures are estimated from model size).

Initial entry. Added 2026-10-28 from the popularity sweep: 14.1k GitHub stars. Release date is the Hugging Face repo date (2026-01-23); the lab publishes none for v1.5 (the technical report followed on 2026-01-31). The XL (4B) series was released 2026-04-02 per the README news. The README says the project is co-led by ACE Studio and StepFun.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (model card HF BAAI/bge-m3).

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF BAAI/bge-reranker-v2-m3).

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF ZhengPeng7/BiRefNet).

Initial entry. Release date is the arXiv paper date (2024-01-07; published in CAAI AIR 2024). MIT licence. Added 2026-10-28 from the popularity sweep: 825k Hugging Face downloads in the last month, 4.3k GitHub stars, and wrapped by rembg (25.2k GitHub stars).

Added hardware requirements (Mac support, memory, NVIDIA). Source: GitHub README + mlx-community/Chatterbox-TTS-4bit + size.

Initial entry. Added 2026-10-28 from the popularity sweep: 27.4k GitHub stars, 1.98M Hugging Face downloads/month. Released date is the Multilingual V3 release (GitHub commit 'v3 multilingual and single language pack release', 2026-06-10). Turbo arrived 2025-12-15; Nano code landed 2026-07-21 (its Hugging Face repo dates from 2026-04-14). The README itself gives no dates.

Deprecated — superseded by Claude Haiku 5.5 (released 2026-10-07). Haiku 4.5 moves to Anthropic's legacy list; still available and listed Active, though its retirement-not-sooner-than date (2026-10-15) has passed.

Initial entry. Released 2026-10-07, superseding Claude Haiku 4.5; $0.10/$0.50 per MTok for prompts up to 100k tokens ($0.50/$2.50 above), 1M-token context, 128k output.

Deprecated — superseded by Claude Opus 5.5 (released 2026-09-22). Opus 5 moves to Anthropic's legacy list; still available, retirement not sooner than 2027-07-24.

Initial entry. Released 2026-09-22 as the Opus-tier flagship, superseding Claude Opus 5; $4/$20 per MTok, 1M-token context, 128k output (300k on Batch with the beta header).

Deprecated — superseded by Claude Sonnet 5.5 (released 2026-09-28). Sonnet 5 moves to Anthropic's legacy list; still available, retirement not sooner than 2027-06-30.

Initial entry. Released 2026-09-28, superseding Claude Sonnet 5; $2/$10 per MTok, 1M-token context, 128k output (300k on Batch with the beta header).

Verified current against Mistral's models overview (v25.08, still listed as active) and changelog; no changes since the last check.

Updated from Embed v4 to Embed 5 (released 2026-09-30): embed-v5.0-pro and embed-v5.0-fast, 128K context, dimensions 256-2048 (default 2048), text/image/mixed inputs, 100+ languages; API price per Cohere's announcement $0.12 (Pro) / $0.08 (Fast) per 1M text tokens. embed-v4.0 is still listed on Cohere's models page. AWS Bedrock does not list Embed 5 yet; the announcement names Microsoft Foundry and Amazon SageMaker.

Updated from Rerank v4 to Rerank 5 (released 2026-10-21): rerank-v5.0-pro and rerank-v5.0-fast, 32K context, Context Focus and calibrated relevance scores; API price per Cohere's announcement $0.05 (Pro) / $0.02 (Fast) per 1M tokens. Cohere's models page now marks rerank-v4.0-pro/-fast as previous generation. Also corrected the Bedrock row: AWS's Cohere page lists only Rerank 3.5, so the earlier 'v4 on Bedrock' strength was dropped.

Deprecated — company closed, no direct successor. coqui.ai showed 'Coqui is shutting down' by 2024-01-03 (Wayback Machine: absent on 2023-12-25, present on 2024-01-03; GitHub issue #3488 'next steps after shutdown' opened the same day). The coqui-ai/TTS repo was last pushed 2024-08-16. No superseded_by because no Coqui successor exists; VoxCPM2, Chatterbox and Qwen3-TTS are the current open alternatives. The weights still show 6.1M Hugging Face downloads/month.

Deprecated — superseded by DeepSeek V4.1 Flash (released 2026-09-10). Weights remain downloadable under the DeepSeek License.

Deprecated — superseded by DeepSeek V4.1 Flash (released 2026-09-10), which includes reasoning with effort control. Weights remain downloadable under MIT.

Removed the OpenRouter provider row: openrouter.ai/deepseek/deepseek-v3 now returns 404.

Added hardware requirements (Mac support, memory, NVIDIA). Source: size of the Hugging Face weights and mlx-community/DeepSeek-V4.1-Flash-1017-oQ2e-mtp.

Initial entry. Released 2026-09-10 per DeepSeek's news index; updated 2026-10-17 as DeepSeek-V4.1-Flash-1017 (coding/agentic RL update, live on web, app and API), which this entry covers rather than a separate file. API ID deepseek-flash at $0.15 in / $0.60 out / $0.003 cache-hit per MTok off-peak (double at peak), 1M context, 384k max output. The legacy IDs deepseek-v4-flash and deepseek-v4-flash-vision-exp still work but are served by V4.1-Flash-1017 at the Flash price. Parameter count: 552B backbone per the card; Hugging Face metadata shows ~763B total.

Deprecated — superseded by DeepSeek V4.1 Flash (released 2026-09-10). DeepSeek's pricing page says the V4 Flash models are retired; the legacy deepseek-v4-flash and deepseek-v4-flash-vision-exp IDs still work but now route to DeepSeek-V4.1-Flash-1017.

Added hardware requirements (Mac support, memory, NVIDIA). Source: size of the Hugging Face weights and mlx-community/DeepSeek-V4-Pro-4bit.

Verified current — DeepSeek's pricing page still lists deepseek-v4-pro as DeepSeek-V4-Pro-0813, priced separately from Flash, and no V4.1 Pro has shipped. DeepSeek's own sources disagree on its future: the 2026-09-10 V4.1 Flash news says V4 Pro is being phased out, with deepseek-v4-pro requests routing to V4.1 Flash from 2026-09-14, while the changelog says V4 Pro API service continues after that date. Re-check next refresh.

Added hardware requirements (Mac support, memory, NVIDIA). Source: Mistral model card; Ollama library page for devstral-small-2.

Initial entry. Added 2026-10-28 from the popularity sweep: 182k Hugging Face downloads/month for the main checkpoint, plus community GGUF and MLX builds. Released 2025-12-09 alongside Devstral 2, per Mistral's changelog. Apache 2.0, 256k context, free to self-host. Mistral's models page lists the API version labs-devstral-small-2512 as deprecated 2026-02-27 and retired 2026-03-31 (alternative: Ministral 3 8B); the weights stay downloadable and there is no newer Devstral Small, so the entry is kept current as a self-host model.

Added hardware requirements (Mac support, memory, NVIDIA). Source: HF model card + size.

Verified still current but dormant: the huggingface/distil-whisper GitHub repo was last pushed 2025-01-08 and the newest checkpoints in the distil-whisper org date from March 2025. Still MIT and usable; for newer open English transcription see NVIDIA Parakeet, and for multilingual see Whisper large-v3 Turbo or Qwen3-ASR.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (github docling-project/docling README (MPS/MLX section)).

Initial entry. First public release v0.1.1 on 2024-07-15 (GitHub releases); latest is v2.143.1 (2026-10-27). MIT licence. Added 2026-10-28 from the popularity sweep: 69.7k GitHub stars.

Updated from Eleven v3 to Eleven v4 (released 2026-09-28): eleven_v4 (10,000-char requests, Text to Dialogue API) plus eleven_v4_turbo (real-time, ~100ms median inference, Text to Dialogue WebSocket), 90+ languages, better voice cloning. ElevenLabs' models page now marks eleven_v3 as previous generation; on 2026-10-05 the ElevenAgents default moved to eleven_v4_turbo.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF google/embeddinggemma-2 card).

Initial entry. Added 2026-10-28 from the popularity sweep: the #1 trending feature-extraction model on Hugging Face, 2,982 likes and 257k downloads/month plus GGUF and ONNX builds. Released 2026-10-06 per Google's Gemma release log (the Hugging Face repo was created 2026-09-14). Apache 2.0, 8,192-token context, 768-dimension output, free to self-host.

Deprecated — superseded by FLUX 3 Image Klein (released 2026-10-15). FLUX.2 [dev] remains downloadable under the FLUX.2 Non-Commercial License; FLUX.2 [klein] 4B (Apache 2.0) noted as the fully permissive small option. Release date corrected from 2026-02-17 (a Hugging Face lastModified date) to 2025-11-25, the launch date on Black Forest Labs' FLUX.2 blog post.

Added hardware requirements (Mac support, memory, NVIDIA). Source: the maker's figures (8 GB with the INT8 build; 11.2 GB in BF16, 5.6 GB in FP8); Mac memory is estimated.

Initial entry. Released 2026-10-15 under the FLUX Kommunity License (free commercial use under $5M annual revenue). Added from the popularity sweep: the ComfyUI repack had 751k Hugging Face downloads in the last month and the main repo 123k, and both were in the top three trending image-to-image models on Hugging Face on 2026-10-28.

Verified current. On 2026-09-18 Google limited 2.5-model access to projects that have actively used them; the deprecations page says the 2.5 models are not deprecated and will be served until further notice, with no shutdown date. New projects are pointed at Gemini 3.5 Flash-Lite or 3.8 Flash.

Verified current. On 2026-09-18 Google limited 2.5-model access to projects that have actively used them; the deprecations page says the 2.5 models are not deprecated and will be served until further notice, with no shutdown date. New projects are pointed at Gemini 3.5 Flash-Lite or 3.8 Flash.

Deprecated — Google's deprecations page sets a shutdown date of 2026-11-17 and names gemini-3.8-live (GA 2026-09-15) as the replacement.

Deprecated — superseded by Gemini 4 Argon (released 2026-10-13). Still served, with no shutdown date on Google's deprecations page. Release date corrected to 2026-02-19 from Google's changelog and deprecations page; the previous 2026-05-07 was the models page's global 'last updated' stamp.

Verified current against Google's models and deprecations pages — stable, no shutdown date, and Google's recommended alternative to the access-limited 2.5 models.

Verified current against Google's changelog, models and deprecations pages — still the head of the Flash line (it replaced the deprecated Gemini 3.7 Flash on 2026-10-08), with no shutdown date.

Initial entry. GA 2026-10-13 per Google's Gemini API changelog as the new flagship, superseding Gemini 3.1 Pro Preview. 1M-token unified context, $2/$10 per MTok through 2026-11-30, then $4/$20 from 2026-12-01.

Initial entry. GA 2026-08-27 per Google's Gemini API changelog; named on the deprecations page as the replacement for Veo 3.1, 3.1 Fast and 3.1 Lite (shut down 2026-10-22). $1.50 input, $9 text output, $17.50 video output per MTok (about $0.10 per second of 720p); 1M-token context.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; Ollama gemma4 page; mlx-community/gemma-4-31b-it-4bit.

Verified current, and two corrections. Licence: Gemma 4 ships under Apache 2.0 (model card `license: apache-2.0`), not the Gemma Terms of Use that covered earlier Gemma generations — so it now shows under the Open filter, and the 'Gemma Terms restrict some use cases' weakness and 'Gated' note are removed. Release date: Google's Gemma releases page lists 'March 31, 2026 Release of Gemma 4', so 2026-03-31 replaces 2026-04-02.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; Ollama gemma4 page; mlx-community gemma-4 builds.

Initial entry. Added 2026-10-28 from the popularity sweep: 7.43M Hugging Face downloads/month for gemma-4-26B-A4B-it, 4.03M for E4B and 3.06M for E2B. Google's release log dates E2B, E4B, 26B A4B and 31B to 2026-03-31 and the 12B Unified to 2026-06-03. Apache 2.0, 128K context (256K on 12B and 26B A4B), free to self-host. The 31B flagship has its own entry.

Added hardware requirements (Mac support, memory, NVIDIA). Source: size of the Hugging Face weights and mlx-community/GLM-5.3-Flash-4bit.

Verified current — still under Latest Models on Z.ai's pricing page at $0.15 in / $0.03 cached / $0.50 out per MTok, matching the list price recorded here. No newer GLM Flash model is listed.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (github zai-org/GLM-OCR README (mlx-vlm, Ollama)).

Initial entry. Release date is the Hugging Face repo date (2026-01-30); the lab publishes none. Model under MIT, SDK code under Apache 2.0. Added 2026-10-28 from the popularity sweep: 1.75M Hugging Face downloads in the last month, 2.2k likes, 7.5k GitHub stars.

Deprecated — superseded by GPT-6 Luna. OpenAI announced deprecation 2026-10-14; lab shutdown 2027-01-14.

Deprecated — superseded by GPT-6 Luna. OpenAI announced deprecation 2026-10-14; lab shutdown 2027-01-14.

Deprecated — superseded by GPT-6 Luna. OpenAI announced deprecation 2026-10-14; lab shutdown 2027-01-14.

Deprecated — OpenAI announced deprecation 2026-10-01, shutdown 2027-01-06; lab replacement gpt-realtime-2.1-mini (not catalogued).

Deprecated — superseded by GPT-6.1 Sol. OpenAI announced deprecation 2026-10-01 (lab replacement gpt-6-sol); lab shutdown 2027-04-01.

Deprecated — superseded by GPT-6 Luna (released 2026-09-22). Still served; not in OpenAI's deprecation table.

Deprecated — superseded by GPT-6.1 Sol (released 2026-09-29). Still served; not in OpenAI's deprecation table.

Deprecated — superseded by GPT-6 Luna (released 2026-09-22). Still served; not in OpenAI's deprecation table.

Deprecated — superseded by GPT-6.1 Sol (released 2026-09-29). Still served; not in OpenAI's deprecation table.

Deprecated — superseded by GPT-6.1 Terra (released 2026-10-26). Still served; not in OpenAI's deprecation table.

Initial entry. Released 2026-09-29 at $2 input / $10 output per MTok (prompts up to 272k input tokens), with a 1.05M-token context and 128k max output. Supersedes GPT-6 Sol (2026-09-22, never catalogued separately), which no longer appears on OpenAI's models overview.

Initial entry. Released 2026-10-26 per OpenAI's API changelog; $1 input / $5 output per MTok, 1.05M-token context and 128k max output per the gpt-6.1-terra model docs.

Verified current against OpenAI's models overview — still listed as a flagship model.

Initial entry. Released 2026-09-22 alongside GPT-6 Sol at $0.10 input / $0.50 output per MTok, with a 1.05M-token context and 128k max output. Named on 2026-10-14 as the replacement for GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini (shutdown 2027-01-14).

Deprecated — superseded by GPT Image 2.5 (released 2026-09-08). Still served; not in OpenAI's deprecation table.

Initial entry. Released 2026-09-08 as two models, Sunburst and Flare (snapshots gpt-image-2.5-sunburst-2026-09-08 and gpt-image-2.5-flare-2026-09-08), at GPT Image 2 token rates with new xhigh and max quality settings.

Verified current against OpenAI's models overview — still listed under Transcription.

Added hardware requirements (Mac support, memory, NVIDIA). Source: OpenAI model card / github.com/openai/gpt-oss; Ollama gpt-oss page (14 GB and 65 GB).

Initial entry. Added 2026-10-28 from the popularity sweep: 20.5k GitHub stars, 5.89M Hugging Face downloads/month for gpt-oss-20b and 2.86M for gpt-oss-120b. Announced by OpenAI 2025-08-05 (Hugging Face repos created 2025-08-04). Apache 2.0, 131,072-token context, free to self-host — no API price applies. Both sizes share this entry; 20b is the headline because it runs on ordinary hardware.

Verified current against OpenAI's models overview — still listed under Transcription.

Deprecated — superseded by Grok 4.7 (released 2026-09-21). Grok 4.6 is still listed on SpaceXAI's models page at the same $2 / $0.50 / $6 per MTok; no retirement date announced.

Initial entry. Released September 2026 per SpaceXAI's release notes; the day (2026-09-21) is from Epoch AI's model database, as SpaceXAI gives only the month. API ID grok-4.7, 500k context, text and image in, text out. $2 in / $0.50 cached / $6 out per MTok under 200k prompt tokens, $4 / $1 / $12 for the whole request at 200k or more. Reasoning effort low / medium / high (default) / xhigh; knowledge cutoff May 2026. Supersedes Grok 4.6.

Verified current — still listed at $0.04 per image. SpaceXAI's release notes say grok-imagine-image-quality retires 2026-11-02, after which its requests are served by grok-imagine-image-2.0 at quality low, at a lower per-image price. grok-imagine-image (1.0) is unaffected.

Added hardware requirements (Mac support, memory, NVIDIA). Source: size of the Hugging Face weights and mlx-community/Kimi-K2.7-Code-mlx-DQ3_K_M-q8.

Initial entry. Added 2026-10-28 from the popularity sweep: 1,425 Hugging Face likes and 90.6k downloads/month. Release date 2026-06-12 is from press coverage of Moonshot's launch — the model card gives no date and the Hugging Face repo was created 2026-06-11. Modified MIT licence (branding rule above 100M monthly users or $20M monthly revenue), 256K context; API pricing not recorded here.

Added hardware requirements (Mac support, memory, NVIDIA). Source: size of the Hugging Face weights.

Verified current — Moonshot's pricing docs still list Kimi K3 with a 1,048,576-token context, and no deprecation notice applies to it (the K2 series closes to new users 2026-10-31). Moonshot's docs now also list a Kimi K3.1 Pro above K3; it is not yet evaluated for the catalog.

Added hardware requirements (Mac support, memory, NVIDIA). Source: https://github.com/hexgrad/kokoro + mlx-community/Kokoro-82M-bf16.

Initial entry. Added 2026-10-28 from the popularity sweep: the top text-to-speech model on Hugging Face by likes (7,292) and downloads (10.6M/month); 9.7k GitHub stars. v1.0 published 2025-01-27 per the model card's release table, following v0.19 (2024-12-25).

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; Ollama llama3.1 page (4.9 GB); mlx-community Meta-Llama-3.1-8B-Instruct-4bit.

Not deprecated: Meta has released no newer small Llama — its latest models are Llama 4 Scout and Maverick (April 2025), with no 8B-class size. For local use it is now outclassed by Gemma 4 E4B / 12B and Qwen3.5 9B, which add image input and thinking mode under the more permissive Apache 2.0.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; Ollama llama3.2-vision page; mlx-community Llama-3.2-11B-Vision-Instruct-4bit.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from 70B size at 4-bit.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from 400B size at 4-bit.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from 109B size at 4-bit.

Added hardware requirements (Mac support, memory, NVIDIA). Source: Lightricks' 80 GB target (32 GB for the distilled quantised build) and the community MLX port's measured memory; Mac minimum is conservative.

Initial entry. Released 2026-08-11 (LTX newsroom, 'Introducing LTX-2.5'), under the LTX-2.x Community License Agreement — free commercial use under $10M annual revenue. Added 2026-10-28 from the popularity sweep: 2.16M Hugging Face downloads in the last month, 9.6k likes, 9.9k GitHub stars, and the top trending image-to-video model on Hugging Face.

Verified current against lumalabs.ai/llm-info, which still names Ray3.2 (June 2026) as the current video model. No Ray4 or newer Ray release.

Added hardware requirements (Mac support, memory, NVIDIA). Source: model card / docs (https://github.com/opendatalab/MinerU (README hardware tier table)).

Initial entry. MinerU 4.0.0 released 2026-09-16 (GitHub release); latest is 4.0.12 (2026-10-11). Licence is the MinerU Open Source License — Apache 2.0 with a commercial-licence threshold and an attribution rule for online services. Added 2026-10-28 from the popularity sweep: 82.7k GitHub stars; the parsing model had 204k Hugging Face downloads in the last month.

Added hardware requirements (Mac support, memory, NVIDIA). Source: MiniCPM-o 4.5 Hugging Face model card (deployment section).

Initial entry. Added 2026-10-28 from the popularity sweep: 27k GitHub stars (the OpenBMB/MiniCPM-o repo now redirects to OpenBMB/MiniCPM-V, which hosts both lines), 280k Hugging Face downloads/month. Released 2026-02-03 per the README news at the time ('We open-source MiniCPM-o 4.5'); API access followed on 2026-05-17.

Added hardware requirements (Mac support, memory, NVIDIA). Source: an estimate from the 33B model size (66 GB of weights) and the maker's 4-GPU serving examples.

Initial entry. Released 2026-07-31 (MiniMax blog) under the MiniMax H3 Community License Agreement, which excludes the US, EU, UK and South Korea, requires a visible "MiniMax H3" credit in commercial products and written authorisation above $20M yearly revenue. Added 2026-10-28 from the popularity sweep: 5.3M Hugging Face downloads in the last month, 6.4k likes, 10.8k GitHub stars.

Initial entry — first MiniMax model in the catalog. Released 2026-10-19. API ID MiniMax-M3.1-Flash at $0.15 in / $0.60 out / $0.015 cached per MTok; 1M context per MiniMax's model overview (the launch blog does not state it). API-only: MiniMax says weights will be released later, so licence is recorded as proprietary until they ship.

Added hardware requirements (Mac support, memory, NVIDIA). Source: the model card, which states 'Inference requires CUDA'.

Initial entry. Added 2026-10-28 from the popularity sweep: 1,484 Hugging Face likes, 992 GitHub stars. Released 2026-08-13 per the MiniMax blog post. Licence terms read from the LICENSE file (last revised August 2026): commercial use allowed if 'MiniMax-Music3' is displayed in the product UI, with written permission needed above US$20M yearly revenue. Global LLM is fine-tuned from Qwen3-8B.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; Ollama ministral-3 page (9.1 GB); mlx-community Ministral-3-14B-Instruct-2512-4bit.

Verified current against Mistral's models overview (v25.12, still listed as active) and changelog. Licence corrected from "Open weights" to Apache 2.0 (as the overview states), which moves it under the Open filter.

Deprecated — superseded by Mistral Large 4 (released 2026-10-06). Still listed as active by Mistral, no retirement date. Licence corrected from 'Mistral Research License' to Apache 2.0 (open, runs locally), per Mistral's model card and the Hugging Face repos; 256k context added from the model card.

Initial entry. Released 2026-10-06 in Public Preview per Mistral's changelog (version 26.10), superseding Mistral Large 3. 1M-token context, 1.05T total / 52B active parameters, $1.36/$4.18 per MTok on the model card; open weights announced as coming soon, so listed as Proprietary until they are published.

Added hardware requirements (Mac support, memory, NVIDIA). Source: size of mlx-community/Mistral-Medium-3.5-128B-4bit; Mistral model card lists llama.cpp and Ollama.

Verified current against Mistral's models overview (v26.04, Modified MIT, still listed as active) and changelog; no changes since the last check.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from 119B size at 4-bit; Mistral model card lists llama.cpp builds.

Verified current against Mistral's models overview (v26.03, still listed as active) and changelog. Licence corrected from "Open weights" to Apache 2.0 (as the overview states), which moves it under the Open filter.

Added hardware requirements (Mac support, memory, NVIDIA). Source: Meta model card meta-models/Muse-Glimmer-30B (24/32 GB targets).

Verified current — Meta's launch post still lists Muse Glimmer as its open-weight model under Apache 2.0. No newer Glimmer version; the other Meta models the post lists (Muse Spark 1.1 and 1.2, Muse Image) are separate lines.

Initial entry. GA 2026-10-06 per Google's Gemini API changelog, deprecating gemini-3.1-flash-image (no shutdown date announced). $1.50 input, $7.50 text output, $30 image output per MTok — $0.0336 / $0.0504 / $0.113 per 1K / 2K / 4K image; 131k-token input.

Added hardware requirements (Mac support, memory, NVIDIA). Source: Hugging Face card nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF (GPU figures) and file size.

Initial entry — first NVIDIA language model in the catalog. Released 2026-10-15 alongside its parent, the open 120B-A12B Nemotron 3.5 Super VL (catalogued here only as context, not as its own entry). 74B total / 9B active MoE in a 24GB native 2-bit pack, 256k context, text/image/video in. Licence OpenMDW-1.1, a permissive model licence (free use, notice-retention, patent-retaliation clause) filed as open-weights because it is not an OSI-approved licence. Hugging Face metadata shows '22B params', which conflicts with the card's own 74B-A9B and likely counts the 2-bit packed tensors; the card's figure is used.

Added hardware requirements (Mac support, memory, NVIDIA). Source: HF model card + mlx-community/parakeet-tdt-0.6b-v3 + size.

Verified the Parakeet entries are unchanged. Newer NVIDIA streaming option: nvidia/nemotron-3.5-asr-streaming-0.6b, released 2026-06-04 per its model card, under the OpenMDW-1.1 licence (not CC BY 4.0), with about 978k Hugging Face downloads/month. Not added as a separate entry here.

Verified current — generally available since 2026-08-31 per Mistral's changelog and still listed as active. OCR 4.0 was deprecated on 2026-09-29 and retired 2026-09-30, leaving OCR 4.1 as the current OCR model.

Verified current against the text-embedding-3-large model docs — live, no deprecation notice, $0.13 per MTok.

Verified current against the text-embedding-3-small model docs — live, no deprecation notice, $0.02 per MTok.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (github PaddlePaddle/PaddleOCR README (mac/linux/win, CPU, M4 speed-up note)).

Initial entry. PaddleOCR 3.7.0 released 2026-06-11 (GitHub release) with PP-OCRv6 and PaddleOCR-VL-1.6, under Apache 2.0. Added 2026-10-28 from the popularity sweep: 91.8k GitHub stars.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; Ollama phi4 page (9.1 GB); mlx-community/phi-4-4bit.

Not deprecated: Microsoft has released no Phi-5, only Phi-4 variants (reasoning, mini, multimodal, and Phi-4-reasoning-vision-15B in January 2026). For local use it is now outclassed by Gemma 4 E4B / 12B and Qwen3.5 9B, which offer 128K-262K context against Phi-4's 16K, plus image input.

Added hardware requirements (Mac support, memory, NVIDIA). Source: https://github.com/OHF-Voice/piper1-gpl + rhasspy/piper-voices.

Licence corrected from MIT to GPL-3.0. The MIT licence belonged to the original rhasspy/piper repo, which is archived (read-only); the only maintained code, OHF-Voice/piper1-gpl, is GPL-3.0 (confirmed via the GitHub API). The rhasspy/piper-voices files on Hugging Face remain MIT. Still active: v1.8.0 (2026-09-04) is the latest release and the repo was pushed 2026-10-25.

Deprecated — superseded by Qwen3.8 27B (released August 2026, recorded as 2026-08-01 because Alibaba publishes only the month). Weights remain on Hugging Face under the Qwen License.

Deprecated — superseded by Qwen3.5 9B / 4B (released 2026-03-02). Weights remain on Hugging Face under Apache 2.0.

Deprecated — superseded by Qwen3-Coder-Next (released 2026-02-03). Weights remain on Hugging Face under Apache 2.0; no retirement date applies to self-hosted weights.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; mlx-community/Qwen3.6-35B-A3B-4bit; Ollama qwen3.6 page.

Un-deprecated and licence corrected. The 2026-09-08 supersession by Qwen3.8 27B was wrong: Qwen3.8 ships a 27B dense model, a 125B-A6B Flash-Next MoE and a 2.4T MoE, so nothing replaces a 35B-A3B MoE, and Qwen3.6-35B-A3B is still heavily used (4.21M Hugging Face downloads/month, 3,044 likes). The licence was also wrong: the Hugging Face model card and LICENSE file are Apache 2.0, not a Qwen License with a 100M-MAU cap, so license_type is now open. Context window (262,144 native, ~1.01M with YaRN) added from the card.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; Ollama qwen3.5 page (9B 6.6 GB); mlx-community/Qwen3.5-9B-MLX-4bit.

Initial entry. Added 2026-10-28 from the popularity sweep: 6.18M Hugging Face downloads/month for Qwen3.5-9B and 7.41M for Qwen3.5-4B. Release date 2026-03-02 is from press coverage of the Qwen team's small-model launch — the model card gives no date and the Hugging Face repos were created 2026-02-27. Apache 2.0, 262,144-token context, free to self-host.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size; mlx-community/Qwen3.8-27B-4bit; Ollama qwen3.8 page.

Verified current — the Hugging Face model card is live under Apache 2.0 with no successor or deprecation notice. Dropped the 'unlike the Qwen3.6 licence' comparison: Qwen3.6 35B-A3B is also Apache 2.0, so the contrast was false.

Added hardware requirements (Mac support, memory, NVIDIA). Source: an estimate from the 20B size and the community mlx-community 8-bit build; the model card states no minimum.

Initial entry. Released 2025-12-23 per the Qwen-Image GitHub news, under Apache 2.0. Added 2026-10-28 from the popularity sweep: 200k Hugging Face downloads in the last month and 1.5k likes. Qwen-Image-2.1 is newer but under the non-commercial Qwen Research License; Qwen-Image-2512 (2025-12-31, Apache 2.0) is the text-to-image sibling, not an editor.

Added hardware requirements (Mac support, memory, NVIDIA). Source: GitHub README + mlx-community/Qwen3-ASR-1.7B-4bit + size.

Initial entry. Added 2026-10-28 from the popularity sweep: 1.39M Hugging Face downloads/month for the 1.7B model; 3.8k GitHub stars. Released 2026-01-29 per the GitHub README news.

Added hardware requirements (Mac support, memory, NVIDIA). Source: size of mlx-community/Qwen3-Coder-Next-4bit.

Initial entry. Added 2026-10-28 from the popularity sweep: 16.8k GitHub stars (QwenLM/Qwen3-Coder), 465k Hugging Face downloads/month plus 813k for the FP8 build. Release date 2026-02-03 is from press coverage of Alibaba's announcement — the model card gives no date and the Hugging Face repo was created 2026-01-30. Apache 2.0, 262,144-token context, free to self-host.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF Qwen/Qwen3-Embedding-0.6B).

Initial entry. Added 2026-10-28 from the popularity sweep: 8.27M Hugging Face downloads/month for Qwen3-Embedding-0.6B. Released 2025-06-05 per the Qwen blog and the arXiv paper. Apache 2.0, 32K context, free to self-host. The multimodal Qwen3-VL-Embedding (Hugging Face repos from January 2026) is a sibling, not covered by this entry.

Added hardware requirements (Mac support, memory, NVIDIA). Source: Qwen3-Omni GitHub README (79 GB BF16 table); mlx-community/Qwen3-Omni-30B-A3B-Instruct-4bit; ggml-org GGUF; llama.cpp issue 21956 (no audio output yet).

Initial entry. Added 2026-10-28 from the popularity sweep: 646k Hugging Face downloads/month for the Instruct model; 4.1k GitHub stars. Released 2025-09-22 per the GitHub README news. Licence is Apache 2.0 (GitHub LICENSE; the Hugging Face card lists it as 'other' with licence name apache-2.0).

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF Qwen/Qwen3-Reranker-4B).

Initial entry. Added 2026-10-28 from the popularity sweep: 1.50M Hugging Face downloads/month for Qwen3-Reranker-0.6B and 2.25M for 4B. Released 2025-06-05 with Qwen3-Embedding, per the Qwen blog and the arXiv paper. Apache 2.0, 32K context, free to self-host.

Added hardware requirements (Mac support, memory, NVIDIA). Source: GitHub README + mlx-community/Qwen3-TTS-12Hz-1.7B-Base-8bit + size.

Initial entry. Added 2026-10-28 from the popularity sweep: 14.1k GitHub stars, 1.88M Hugging Face downloads/month for the 1.7B CustomVoice model. Released 2026-01-22 per the GitHub README news; no newer Qwen TTS release on Hugging Face.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (github xinntao/Real-ESRGAN README (ncnn-vulkan portable builds)).

Still current, not deprecated: the repo is dormant (last commit 2024-04, last push 2024-08) but Real-ESRGAN is still the engine inside Upscayl (51.2k GitHub stars, AGPL-3.0), the popular free desktop upscaler, added as a provider. Newer option for higher detail: SeedVR2.

Verified current against Runway's API model docs: Gen-4.5 is still the newest Gen model and no Gen-5 exists. Aleph 2.0 (2026-06-02) is Runway's newer flagship but is an in-context video editing model, not a Gen successor. Swapped Sora 2 out of the rivals weakness (OpenAI shut it down 2026-09-24) for Kling.

Added hardware requirements (Mac support, memory, NVIDIA). Source: an estimate from the 3B/7B model sizes and the community Mac builds; the lab's own video test used one 80 GB H100.

Initial entry. Release date is the arXiv paper date (2025-06-05); weights and code followed on 2025-06-10/11. Apache 2.0. Added 2026-10-28 from the popularity sweep: the two main ComfyUI repacks (numz and Comfy-Org) had about 1.4M Hugging Face downloads combined in the last month; official repo 1.4k GitHub stars.

Added hardware requirements (Mac support, memory, NVIDIA). Source: the Stability stable-audio-3 README (peak memory about 6.5 GB for medium; Mac memory is estimated).

Verified current against Stability's news page. No Stable Audio release newer than 3.0; the 2026-08-18 post adds new ways to use 3.0 (a DAW plugin and a more advanced generation experience), not a new model.

Added hardware requirements (Mac support, memory, NVIDIA). Source: an estimate from the model sizes (Medium about 2.5B, Large about 8B).

Verified current against Stability's news page: its 2026 posts announce no Stable Diffusion image model after 3.5.

Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF stabilityai/stable-diffusion-x4-upscaler card (attention slicing for low VRAM)).

Deprecated with no successor — BigCode has published no model since April 2024, and current open coding models (Qwen3-Coder-Next, Devstral Small 2) far outclass it with 256K context against 16K. Dated to this review because no lab deprecation or successor release exists to date it by.

Updated from v5.5 to v6 (released 2026-09-09) per suno.com/blog/introducing-v6. This is the industry-partnered generation the previous watch item anticipated: Suno says v6 was 'developed with' Warner Music Group, BMG and Believe (preceded by the BMG partnership post of 2026-08-12 and the Believe/TuneCore post of 2026-09-08). Ships as v6 and v6-wild for Pro/Premier and v6-mini for everyone. Suno says it will retire previous models as v6 rolls out, but gives no date for v5.5. Removed the 'Suno API' provider row: suno.com/developers returns 404 and no public API page was found.

Deprecated — no direct successor from the authors. 2021 model, repo dormant since 2024-05 (last code commit 2022-12-04). Use SeedVR2 or Real-ESRGAN instead.

Added hardware requirements (Mac support, memory, NVIDIA). Source: model card / docs (https://docs.topazlabs.com/gigapixel-ai/system-requirements (Topaz docs, legacy page for v8)).

Deprecated — superseded by Gemini Omni Flash (released 2026-08-27). Shut down: veo-3.1-generate-preview and veo-3.1-fast-generate-preview were turned off on 2026-10-22, per Google's deprecations page.

Deprecated — superseded by Gemini Omni Flash (released 2026-08-27). Shut down: veo-3.1-lite-generate-preview was turned off on 2026-10-22, per Google's deprecations page.

Added hardware requirements (Mac support, memory, NVIDIA). Source: https://github.com/microsoft/VibeASR.cpp + size.

Initial entry. Added 2026-10-28 from the popularity sweep: 55k GitHub stars (whole VibeVoice repo), 677k Hugging Face downloads/month. Open-sourced 2026-01-21 per the README news; BitNet CPU build 2026-07-23; streaming 7B model 2026-09-03. The README also records Microsoft removing the VibeVoice-TTS code on 2025-09-05 over misuse.

Added hardware requirements (Mac support, memory, NVIDIA). Source: https://alphacephei.com/vosk/models.

Added hardware requirements (Mac support, memory, NVIDIA). Source: https://github.com/OpenBMB/VoxCPM.

Initial entry. Added 2026-10-28 from the popularity sweep: 40.8k GitHub stars, 983k Hugging Face downloads/month, 1,732 likes. The README dates VoxCPM2 only to April 2026; the release date used is the 2.0.0 GitHub release tag (2026-04-06). Replaces VoxCPM-0.5B (2025-09) and VoxCPM1.5 (2025-12), which supported only Chinese and English.

Added hardware requirements (Mac support, memory, NVIDIA). Source: https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602.

Initial entry. Added 2026-10-28 from the popularity sweep: 2.03M Hugging Face downloads/month. Released 2026-02-04 per Mistral's changelog (alongside Voxtral Mini Transcribe 2). Mistral released the Voxtral 3 audio family on 2026-10-20, including Voxtral Mini Transcribe Realtime 3, as API public previews; this open-weight 2602 model remains the current downloadable one.

Verified current against Voyage's embeddings docs. The 4 series (voyage-4-large / voyage-4 / voyage-4-lite) is still the current general-purpose family; voyage-3.5 and voyage-3-large sit under 'Older models'. No voyage-5 exists.

Added hardware requirements (Mac support, memory, NVIDIA). Source: the Wan2.2 README (5B model on 24 GB, 14B models on 80 GB); Mac memory is estimated.

Initial entry. Released 2025-07-28 per the GitHub README news, under Apache 2.0. Added 2026-10-28 from the popularity sweep: 18.2k GitHub stars, and the Comfy-Org ComfyUI repack had 8.5M Hugging Face downloads in the last month — the most-downloaded image-to-video repo on Hugging Face. Still the newest open Wan line — no later Wan version has open weights under Wan-AI on Hugging Face; the most recent upload is Wan2.2-Animate-2-14B (2026-07).

Added hardware requirements (Mac support, memory, NVIDIA). Source: https://github.com/openai/whisper (about 10 GB VRAM for large) + mlx-community/whisper-large-v3-mlx.

Added hardware requirements (Mac support, memory, NVIDIA). Source: https://github.com/openai/whisper (about 6 GB VRAM for turbo) + mlx-community/whisper-large-v3-turbo.

Added hardware requirements (Mac support, memory, NVIDIA). Source: the model card and README (16 GB graphics card; 4 GB with stable-diffusion.cpp); Mac memory is estimated.

Initial entry. Released 2025-11-26 per the GitHub README news, under Apache 2.0. Added 2026-10-28 from the popularity sweep: 12.1k GitHub stars, 730k Hugging Face downloads in the last month and 5.5k likes.

Model names refreshed to the September 2026 lineup. The benchmark figures deliberately keep the model they were measured on, because those tests have not been re-run on the newer models; they are now dated so the age is visible.

Cost rows updated to Claude Sonnet 5 ($2/$10) and GPT-5.6 Terra ($2/$12), replacing Sonnet 4.6 and the deprecated GPT-4o. The Sonnet per-character estimate only fell to ~$3.50 rather than by the full price cut, because Claude 5's tokenizer produces roughly 30% more tokens for the same text.

Verified against claude.com/pricing. Corrected the Claude Team price — it is $20/seat annually, not $25 — and added the Team Premium tier at $100/seat. Updated the model-under-the-hood row from Sonnet 4.6 / Opus 4.7 to Sonnet 5 / Opus 5, noting that Opus requires a paid plan. Jasper, Copy.ai and Writer pricing was not re-verified this round.

Model names in the inference-cost row updated to Claude Sonnet 5 and GPT-5.6 Terra. GPT-4o was named here and is deprecated, superseded by GPT-5.5. The cost range is unchanged. Added referenced-model links for models this page already names: Claude Sonnet 5, GPT-5.6 Terra.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Two corrections. Claude's context window is set by the model, not the plan: any paid plan gets 1M when chatting with Opus 5, Sonnet 5 or Fable 5.1, and older models give 500K or 200K. And Microsoft names no model at all, routing automatically per prompt. Flagship models updated to Claude Opus 5 and Sonnet 5.

Corrected the model prerequisite. It described Claude as '200k default, 1M on the higher tiers', which is backwards for the current generation: every Claude 5 model is 1M with no smaller variant to fall back to. The deprecated GPT-4o reference was replaced.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Model and plan rows re-verified against each vendor's own docs. Cursor does not offer GPT-6 Astra; GitHub Copilot has added a Max tier at $100/month; Claude Code defaults to Opus 5 on Max and above and Sonnet 5 on Pro. Claude Pro corrected to $17/month billed annually.

Recommended default models updated from Claude Sonnet 4.6 and GPT-4o to Claude Sonnet 5 and GPT-5.6 Terra, in both the prose and the cost row. GPT-4o is deprecated and two generations behind, so the old recommendation was steering readers at a model OpenAI has superseded. Added referenced-model links for models this page already names: Claude Sonnet 5, GPT-5.6 Terra.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Updated the Claude Vision reference from Sonnet 4.6 / Opus 4.7 to Sonnet 5 / Opus 5. Repointed the referenced-models links to Mistral OCR 4.1 and the current Claude and GPT entries.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Added referenced-model links for models this page already names: GPT-4.1, OpenAI text-embedding-3-large. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Added referenced-model links for models this page already names: GPT-4.1. Page content not otherwise re-verified.

Added referenced-model links for models this page already names: BGE-M3, Cohere Embed v4, OpenAI text-embedding-3-large, OpenAI text-embedding-3-small. Page content not otherwise re-verified.

Added referenced-model links for models this page already names: OpenAI text-embedding-3-large, OpenAI text-embedding-3-small. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Flagship-model cost row updated to Claude Sonnet 5 / GPT-5.6 Terra / Gemini 3.1, and the referenced models repointed to Mistral OCR 4.1 and the current Claude and GPT entries. The per-page cost range is unchanged.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Claude 4.6 Sonnet to Claude Sonnet 5; GPT-4o to GPT-5.6 Terra. Added referenced-model links for models this page already names: OpenAI text-embedding-3-small. Page content not otherwise re-verified.

Updated the per-image vision cost row to Claude Sonnet 5 / GPT-5.6 Terra / Gemini 3.1. The cost range is unchanged. Referenced-model links updated to the current lineup: GPT-4o to GPT-5.6 Terra; Claude 4.7 Opus to Claude Opus 5. Added referenced-model links for models this page already names: Claude Sonnet 5, GPT-5.5.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Pricing, context windows and model names refreshed to the September 2026 lineup. The quality rankings still come from the Q1 2026 blind evaluation of the versions current at the May 2026 snapshot; that has not been re-run, and the page now says so up front.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Added referenced-model links for models this page already names: Gemini 2.5 Flash. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Frontier and open-weight model references refreshed to the September 2026 lineup. The open-weight row is now tiered by hardware, because naming DeepSeek V4 Pro flat in the 'Local' column implied a 1.6T model runs on the workstation budgeted two rows below.

Referenced-model links updated to the current lineup: Claude 4.7 Opus to Claude Opus 5; GPT-4o to GPT-5.6 Terra. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Claude 4.6 Sonnet to Claude Sonnet 5. Page content not otherwise re-verified.

Updated the proprietary-flagship price stat to the September 2026 lineup (GPT-6 Astra, Claude Opus 5, Gemini 3.1 Pro). The range moved from $3–$15 to $2–$10 per million input tokens. Added referenced-model links for models this page already names: Claude Haiku 4.5, Claude Opus 5, Gemini 3.5 Flash-Lite, GPT-5.6 Luna, GPT-6 Astra.

The model-choice step now leads with Qwen3.8 27B, which is Apache 2.0 and fits the single-GPU pilot tier, instead of Llama 3.3 70B, which pushed readers toward far more expensive hardware than the job needs. The measured latency figures stay with the model they were measured on.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Cost estimate moved from Claude Sonnet 4.6 to Sonnet 5 pricing. The range tightened only slightly (~$0.01-$0.035 from ~$0.01-$0.04) rather than falling by the full third the price cut implies, because Claude 5's tokenizer produces roughly 30% more tokens for the same text. Added referenced-model links for models this page already names: Claude Sonnet 5.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: Claude 4.7 Opus to Claude Opus 5. Page content not otherwise re-verified.

Corrected the model-API prerequisite from 'Claude (200k/1M), GPT (128k-272k)' to the current windows: Claude 5 is 1M on every model, GPT-5.6 and GPT-6 reach ~1.05M. last_verified deliberately not bumped: only this line was re-checked, not the whole page. Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite.

Per-summary cost recalculated at Claude Sonnet 5 pricing: about $0.03, down from $0.05. Claude's context figure corrected too, since any paid plan chatting with Opus 5, Sonnet 5 or Fable 5.1 gets the full 1M.

New section on what actually sets your context window: the model you pick, not the plan you pay for. Any paid Claude plan gets 1M on Opus 5, Sonnet 5 or Fable 5.1, while older models give 500K or 200K. Pricing refreshed to the September 2026 lineup and both worked cost examples recalculated.

Referenced-model links updated to the current lineup: Gemini 3.1 Flash-Lite to Gemini 3.5 Flash-Lite. Added referenced-model links for models this page already names: GPT-4.1. Page content not otherwise re-verified.

Added referenced-model links for models this page already names: BGE-M3. Page content not otherwise re-verified.

Referenced-model links updated to the current lineup: OpenAI TTS to GPT Transcribe. Page content not otherwise re-verified.

Updated the agent-capability example from Claude Sonnet 4.5+/Opus 4.7 to Claude Sonnet 5 / Opus 5. Added referenced-model links for models this page already names: Claude Sonnet 5.

Referenced-model links updated to the current lineup: GPT-4o to GPT-5.6 Terra; Claude 4.7 Opus to Claude Opus 5. Added referenced-model links for models this page already names: GPT-5.5. Page content not otherwise re-verified.

Added referenced-model links for models this page already names: Distil-Whisper. Page content not otherwise re-verified.

Verified current. No bge-m4 exists; BGE-M3's successors are task-specific siblings rather than a new generation.

Verified current. No bge-reranker-v3 exists. The newer v2.5-gemma2-lightweight is a heavier variant under the restrictive Gemma licence, not a general replacement.

Deprecated — superseded by Claude Sonnet 5 (released 2026-06-30), documented by Anthropic as a drop-in replacement. Sonnet 4.6 moves to the legacy table; retirement not sooner than 2027-02-17.

Deprecated — superseded by Claude Fable 5.1 (released 2026-09-01). Fable 5 moves to Anthropic's legacy table; still available, retirement not sooner than 2027-06-09.

Initial entry. Released 2026-09-01 as the successor to Claude Fable 5, at the same $10/$50 per MTok with cache reads cut to $0.25 per MTok. Adds adaptive thinking always on; drops support for forced tool use.

Verified current against Anthropic's models overview. Still a column in the current-lineup table; no Haiku 5 exists. Retirement not sooner than 2026-10-15.

Deprecated — superseded by Claude Mythos 5.1 (released 2026-09-01). Still in limited availability; not listed on Anthropic's deprecation table.

Initial entry. Released 2026-09-01 alongside Claude Fable 5.1, succeeding Claude Mythos 5. Remains invitation-only via Project Glasswing.

Deprecated — superseded by Claude Opus 5 (released 2026-07-24). Opus 4.8 moves to Anthropic's legacy table; still available on all platforms, retirement not sooner than 2027-05-28.

Initial entry. Released 2026-07-24 as the Opus-tier flagship, superseding Claude Opus 4.8. 1M-token context (default and maximum), $5/$25 per MTok, adaptive thinking on by default with a low/medium/high/xhigh/max effort ladder.

Initial entry. Released 2026-06-30 as a drop-in replacement for Claude Sonnet 4.6. The $2/$10 per MTok launch pricing became the standard price — the increase to $3/$15 scheduled for 2026-09-01 was cancelled.

Verified current against Mistral's models overview — the only model in the Code Models section, and not on the deprecation table.

Verified current against Cohere's models page. embed-v4.0 is still the newest Embed generation; everything below is v3.0/v2.0. No v5 exists.

Verified current against Cohere's models page and the Rerank v4.0 changelog. rerank-v4.0-pro is top of the rerank list; rerank-v4.0-fast is a sibling tier, not a successor.

Verified: the weights remain hosted, ungated and heavily used (7.2M downloads last month) despite Coqui the company being defunct — this entry's existing 'community maintenance only' weakness is accurate. Maintenance has moved to idiap/coqui-ai-TTS.

Corrected on three counts. Release date 2026-05-06 was a Hugging Face push timestamp, not a release — the official DeepSeek-V4-Flash-0731 shipped 2026-07-31 per DeepSeek's API changelog, superseding the preview this entry described. Licence corrected from 'DeepSeek License' to MIT, and the false 'restrictions on commercial use cases' weakness removed. Added parameter size and 1M context window.

Corrected on three counts. Release date 2026-05-06 was a Hugging Face push timestamp, not a release — the GA DeepSeek-V4-Pro-0813 shipped 2026-08-13 per DeepSeek's API changelog, superseding the preview this entry described. Licence corrected from 'DeepSeek License' to MIT, and the false 'restrictions on commercial use cases' weakness removed. Added parameter size and 1M context window.

Deprecated — Mistral's deprecation table shows devstral-2512 deprecated 2026-05-22 and retired 2026-07-31, with Mistral Medium 3.5 as the stated alternative. The entry had been showing a retired model as current.

Verified current against the full distil-whisper org (12 models). The distil-large-v3.5 family is newest; no v4 exists. Worth knowing: distil-large-v3 still has roughly 90x the downloads of v3.5.

Dropped Turbo v2.5 from purpose and strengths — ElevenLabs' models page now marks Eleven Turbo v2.5 and Turbo v2 as deprecated, 'outclassed by Flash models', with eleven_flash_v2_5 as the replacement. Eleven v3 itself verified current; no v4 exists.

Verified current against BFL's FLUX.2 launch post. No newer [pro] version exists; FLUX.2 [max] is a sibling tier, not a successor.

Verified current against Black Forest Labs' docs, which state FLUX.2 remains fully supported for production. FLUX 3 exists but its Dev open-weight backbone has not shipped.

Verified current against Google's models and deprecations pages. Still GA and priced; no shutdown announced. Generations behind the Flash line but not superseded within its own 2.5 family.

Verified current against Google's models and deprecations pages. Still GA with no shutdown announced — and now the only GA (non-preview) Pro model, since Gemini 3.5 Pro was shelved.

Deprecated by Google — the deprecations page gives a 2027-05-07 shutdown and names Gemini 3.5 Flash-Lite (2026-07-21) as the replacement.

Release date corrected to 2026-03-11 from Google's deprecations page. The previous 2026-05-07 was the models page's global 'last updated' stamp. Verified still current with no shutdown announced.

Verified current. Still the top Pro model on Google's docs — Gemini 3.5 Pro was announced at I/O, missed three times and reportedly shelved to pretrain Gemini 4. No Pro model newer than this exists.

Deprecated — Google's models page now describes it as the legacy Flash model. Superseded by Gemini 3.8 Flash (2026-09-02), with 3.6 and 3.7 Flash released in between. No shutdown date announced.

Initial entry. Released 2026-07-21; named on Google's deprecations page as the replacement for Gemini 3.1 Flash-Lite (shutdown 2027-05-07). Context window left unset — Google publishes none.

Initial entry. Released 2026-09-02 per Google's Gemini API changelog as the head of the Flash line, superseding Gemini 3.7 Flash (2026-08-13) and 3.6 Flash (2026-07-21). Dates taken from the changelog and deprecations pages, not the models page (whose per-model stamps are a global page timestamp).

Release date corrected to 2025-12-17 from Google's deprecations page; the previous 2026-05-07 was the models page's global 'last updated' stamp. Note Google names gemini-3.6-flash as the official replacement, though the catalog chains it through Gemini 3.5 Flash.

Verified current against Google's Gemma pages. Gemma 4 (April 2026) is the newest generation and 31B the top core size. Gemma is absent from the Gemini API models page by design — that is a coverage gap, not a deprecation.

Initial entry — first Z.ai model in the catalog. Released 2026-08-26, eight days after the flagship GLM-5.3 (2026-08-18). The Flash variant is catalogued rather than the flagship because it is plain MIT, while GLM-5.3 carries a custom licence requiring a Z.ai security review for MaaS operators above $10B revenue. Pricing recorded at list ($0.15/$0.50); a 50% promotional rate expires 2026-09-09. Release date taken from Z.ai's release notes, NOT from the arXiv paper linked on the model card — that paper (2026-02-17) is the shared GLM-5 base-model report, not this model's release.

Verified current against OpenAI's models overview — still listed under speech.

Deprecated by OpenAI — announced 2026-08-26, shutdown 2027-02-26, replaced by gpt-transcribe and gpt-live-transcribe.

Pricing corrected to $0.20 input / $1.20 output per MTok — OpenAI cut Luna's rates about 80% on 2026-07-30; the entry carried the July launch prices.

Pricing corrected to $4 input / $20 output per MTok — OpenAI cut Sol's rates on 2026-08-21 (promotional at least through 2026-11-21); the entry carried the July launch prices. Repositioned below GPT-6 Astra (released 2026-09-03).

Pricing corrected to $2 input / $12 output per MTok — OpenAI cut Terra's rates roughly 20% on 2026-07-30; the entry carried the July launch prices.

Initial entry. Released 2026-09-03 per OpenAI's API changelog as 'our most capable model, built for the hardest end-to-end work'. Specs and pricing verified against the gpt-6-astra model docs.

Verified current against OpenAI's models overview — still listed, snapshot gpt-image-2-2026-04-21.

Initial entry. Released 2026-07-28. Named alongside gpt-transcribe on OpenAI's deprecations page as a replacement for gpt-4o-transcribe (shutdown 2027-02-26). Context window and max output are not published by OpenAI.

Initial entry. Released 2026-07-28. Named on OpenAI's deprecations page as the replacement for gpt-4o-transcribe (shutdown 2027-02-26). Context window and max output are not published by OpenAI, so both are left unset.

Initial entry — first SpaceXAI model in the catalog. Released 2026-08-12. Maker recorded as SpaceXAI: that is the string on the copyright notice, the nav on every announcement page, the model docs' own prose, and the Arena leaderboard's provider column. The company was formerly xAI, and the API and SDK identifiers are still xai_sdk / XAI_API_KEY / api.x.ai. Reasoning controls run low / medium / high / xhigh with no way to disable.

Initial entry. Released 2026-08-07. SpaceXAI's announcement claims #2 worldwide on the Arena text-to-image and image-edit leaderboards as of the launch date; the live text-to-image leaderboard on 2026-09-04 shows it at #3, displaced by Microsoft's mai-image-2.6, with gpt-image-2 still #1. The entry records the current position rather than the launch claim. Current Image Edit standing was not re-checked.

Initial entry — first Moonshot AI model in the catalog. Weights released 2026-07-27; the API launched 2026-07-17. The licence was read in full: contrary to widely repeated secondary reports, it contains NO revenue-sharing clause and no percentage. What it has is a negotiation gate — MaaS operators above $20M revenue over any rolling 12 months must agree separate terms first — plus an attribution requirement above 100M MAU or $20M monthly revenue. Internal use and access through Moonshot's own products are exempt.

Verified current against Kuaishou's own Kling 3.0 quickstart guide. Nothing newer than the 3.0 series is named; all 'Kling 4.0' material is forward-looking speculation.

Release date corrected to 2025-04-05, Meta's actual Llama 4 launch. The previous 2025-05-22 was the Hugging Face repo push timestamp. Still the newest Llama — there is no Llama 5; Meta's open-weight work moved to the Muse family (see Muse Glimmer 30B).

Release date corrected to 2025-04-05, Meta's actual Llama 4 launch. The previous 2025-05-22 was the Hugging Face repo push timestamp. Still the newest Llama — there is no Llama 5; Meta's open-weight work moved to the Muse family (see Muse Glimmer 30B).

Updated from Luma Ray3 to Ray3.2 (released 2026-06-09), confirmed as the current model on lumalabs.ai/llm-info. Ray3.14 shipped 2026-01-26 in between. Note for future runs: docs.lumalabs.ai/docs/video-generation is a generation behind (it documents only the deprecated ray-2 line) — use lumalabs.ai/llm-info as the verification source instead. No Ray4 exists.

Verified current against Mistral's models overview and model card; release date matches exactly.

Verified current against Mistral's models overview and model card; release date matches exactly.

Release date corrected to 2026-04-28 from Mistral's own model card and changelog; the previous 2026-04-01 was inferred from the 26.04 version stamp. Verified current — Mistral now names it the alternative for both the retired Devstral 2 and the retired Magistral line, with reasoning exposed through the reasoning_effort parameter rather than a separate model.

Verified current against Mistral's models overview and model card; release date matches exactly. Now described as a hybrid model unifying instruct, reasoning and coding.

Initial entry. Released 2026-08-10 under the new meta-models Hugging Face org. Meta's open-weight work has moved off the Llama name — there is no Llama 5, and the closed-weight Muse Spark (2026-04-08) replaced Llama as Meta's flagship. Llama 3.1 8B is kept in the catalog because Llama 4's smallest model is 109B, leaving Muse Glimmer as the only current Meta option for a single consumer GPU.

Verified current — no Parakeet v4 exists. Note nvidia/parakeet-unified-en-0.6b (2026) now beats this model both offline and dramatically in streaming, so the 'top-of-leaderboard English accuracy' claim needs a re-check against the current Open ASR leaderboard.

Superseded by OCR 4.1. Release date corrected to 2025-12-18 to match Mistral's model card. Still available and not on Mistral's deprecation table.

Initial entry. Released 2026-07-16 and generally available 2026-08-31 per Mistral's changelog; mistral-ocr-latest and mistral-ocr-4 now point at it. OCR 4.0 shipped nine weeks earlier and is skipped as an interim version.

Verified current against the model card — 14B, 16K context, MIT, no deprecation notice. Phi-5 does not exist.

Pika 2.5 confirmed as the current model — Pika's own developer docs list pika/pika-2.5/text-to-video and image-to-video with published pricing ($0.04/sec at 720p, $0.09/sec at 1080p, 5s text-to-video cap). Entry held at 2.2 because Pika publishes no release date for 2.5 anywhere, and this catalog does not guess release dates. Bump the title once a dated source appears.

Verified actively maintained: OHF-Voice/piper1-gpl shipped v1.8.0 on 2026-09-04, with v1.5.0 through v1.7.0 across July and August 2026. The Open Home Foundation's README notes it is seeking maintainers.

Superseded by Qwen3.8 27B (August 2026). Note the filename is a legacy artifact of the corrected 2026-06-08 entry and no longer matches the title.

Initial entry. Qwen3.8 released August 2026, superseding Qwen3.6 (April 2026) with Qwen3.5 (February 2026) in between. Alibaba's model card gives only a month, so the release date is recorded as the first of the month rather than a guessed day. Benchmarks and licence taken from the model card, not from Hugging Face push timestamps.

Verified current and not archived; last release v0.3.0 (2022-09-20). No successor exists — the README's GFPGAN/BasicSR links are complementary projects, not replacements. Canonical repo remains under the author's personal account; TencentARC/Real-ESRGAN returns 404.

Verified current against Runway's model docs and changelog, which dates Gen-4.5 to 2025-12-11 — matching this entry exactly. No Gen-5 exists. Aggregator claims of a March 2026 launch contradict Runway's own changelog and were rejected.

Verified current against Stability's news page, which dates Stable Audio 3.0 to 2026-05-20 — matching this entry exactly.

Verified current. A full Hugging Face org enumeration by creation date shows no Stable Diffusion 4 repo, and Stability's own 2026 news page never mentions one — the widely circulated 'SD4, April 2026' claim is fabricated.

Verified current — no newer open-weights Stability upscaler exists in the full org listing.

Verified current. No StarCoder3 exists. BigCode still ships evaluation datasets but has published no new model since 2024, so dormancy here is not staleness.

Verified current against suno.com and Suno's release notes, which confirm v5.5 shipped 2026-03-26 — matching this entry exactly. No v6 exists. Watch item: Suno has said current models will be deprecated when its Warner/BMG-licensed models ship later in 2026.

Verified current as the canonical single-image restoration implementation. Repo is not archived; last real commit 2022-12-04. The authors' VRT/RVRT work is video restoration — adjacent, explicitly not a replacement.

Corrected on four counts. Renamed from 'Topaz Gigapixel AI' to 'Topaz Gigapixel' — Topaz dropped the 'AI' suffix. Release date moved from 2018-06-01 to 2025-09-16, the current product's date per the product page's schema markup; the 2018 date predated the current product entirely. Product URL updated from /gigapixel-ai to /gigapixel. Pricing corrected from a one-time licence to subscription-only. Version 1.3.3 shipped 2026-07-27.

Verified current against Udio's own help-centre changelog, which names Allegro v1.5 (2025-03-18) as the newest model — matching this entry exactly. Third-party claims of a 'Udio v4' or 'v3.5' have no official backing anywhere and were rejected again this run.

Release date corrected to 2025-10-15 from Google's deprecations page. The previous 2026-05-07 was the models page's global 'last updated' stamp, not a release date. Verified still current — no Veo 4 exists.

Release date corrected to 2026-03-31 from Google's deprecations page. The previous 2026-05-07 was the models page's global 'last updated' stamp, not a release date. Verified still current.

Verified actively maintained — real feature commits through 2026-08-09, despite no tagged release since v0.3.50 (2024-04-22). The release gap is not staleness. No successor project exists.

Verified current against Voyage's embeddings docs. voyage-4-large is still the best general-purpose retrieval model. Note voyage-context-4 is a different family (contextualised chunk embeddings), not a newer version of this one.

Link fix: OpenAI's image guide moved — /api/docs/guides/images now 404s. Repointed the provider and official_link to /api/docs/guides/image-generation.

Link fix: promoted the models overview (elevenlabs.io/docs/overview/models) to the primary docs link. The API reference was the only docs link, and it does not enumerate models — which made it a poor verification source. Kept as a secondary link.

Verified current after the GPT-5.6 launch — GPT-5.6 is an additional frontier tier, not a replacement. OpenAI's deprecations page does not list GPT-5.5, and still names it as the recommended successor for o3, GPT-5, and the chat-latest snapshots. Release date corrected 2026-04-23 → 2026-04-24 per the API changelog.

Initial entry. GPT-5.6 family released 2026-07-09 per OpenAI's API changelog; specs and pricing verified against the gpt-5.6-luna model docs.

Initial entry. GPT-5.6 family released 2026-07-09 per OpenAI's API changelog; specs and pricing verified against the gpt-5.6-sol model docs. API alias is `gpt-5.6`.

Initial entry. GPT-5.6 family released 2026-07-09 per OpenAI's API changelog; specs and pricing verified against the gpt-5.6-terra model docs.

Deprecated — superseded by Mistral Medium 3.5 per Mistral's own Legacy/Deprecated table (deprecated 2026-05-22, retirement 2026-07-31). Whole Magistral line retired; no successor reasoning model.

Deprecated — superseded by GPT-5.5. OpenAI's deprecations page now lists an API shutdown of 2026-12-11 for o3-2025-04-16 (announced 2026-06-11), reversing the prior 'no API change' position recorded below.

Deprecated — retired with no successor. OpenAI's deprecations page lists all sora-2 aliases and snapshots shutting down 2026-09-24 (announced 2026-03-24) with an empty replacement column. Recorded without superseded_by; the schema refinement was relaxed to permit successor-less retirements.

Updated OpenAI column from gpt-image-1 to gpt-image-2 (released 2026-04-21): category-leading text rendering, native 2K output, LLM-planned rendering, token-based pricing tiers.

Initial entry. Released 2026-06-09 as Anthropic's most capable widely released model — the first publicly available Mythos-class model, a new tier above Opus 4.8.

Initial entry. Released 2026-06-09 in limited availability via Project Glasswing, succeeding Claude Mythos Preview.

Repositioned as Anthropic's most capable Opus-tier model: Claude Fable 5 launched 2026-06-09 as a new tier above Opus. Opus 4.8 is not superseded and remains the standard-lineup flagship.

Verified current vs Anthropic docs. Corrected release date to 2026-02-17 (prior 2025-09-29 was the Sonnet 4.5 snapshot date).

Deprecated — superseded by Claude Opus 4.8; Opus 4.7 is now in Anthropic's legacy table.

Initial entry (2026-06-08 catalog refresh, Tier 3). Current flagship Opus, released 2026-05-28; supersedes Claude 4.7 Opus.

Updated from Embed v3 to embed-v4.0 (released 2025-04-15): multimodal, 128K context. The v3.0 models remain available but are no longer the flagship.

Updated from Rerank v3 to rerank-v4.0-pro (released 2025-12-11). (Interim: rerank-v3.5 2024-12-02.)

Updated to distil-large-v3.5 (trained on ~4× more data; faster than Whisper-large-v3-Turbo on long-form).

Updated from Multilingual v2 to Eleven v3 (GA 2026-02-02), 70+ languages. Multilingual v2 remains available but is no longer the flagship.

Updated from FLUX.1 [pro] to FLUX.2 [pro] (announced 2025-11-25), the current commercial flagship. FLUX.1 [pro] remains callable via the legacy /flux-pro endpoint.

Deprecated — Gemini 2.5 Deep Think is no longer a current API model; Deep Think now lives in the Gemini 3.x generation.

Initial entry (2026-06-08 catalog refresh, Tier 3). GA model released 2026-05-19 at Google I/O; supersedes the Gemini 3 Flash preview.

Deprecated — superseded by the GA Gemini 3.5 Flash.

Verified current vs Google docs. Corrected release date to 2026-04-02 (Gemma 4 launch per Google blog; prior 2026-05-07 was an HF lastModified timestamp).

Deprecated — superseded by GPT-5.5 (OpenAI 'Deprecated' badge; retired from ChatGPT 2026-02-13; oldest snapshot API shutdown 2026-10-23).

Verified current. The base alias now points to the gpt-4o-mini-tts-2025-12-15 snapshot; the 2025-03-20 snapshot retires 2026-07-23.

Initial entry (2026-06-08 catalog refresh, Tier 3). OpenAI's hosted-API STT model; complements the open-weights Whisper entries.

Initial entry (2026-06-08 catalog refresh, Tier 3). Released 2026-02-05; OpenAI's flagship agentic coding model.

Initial entry (2026-06-08 catalog refresh, Tier 3). Released 2026-03-17; OpenAI's named replacement for o4-mini.

Initial entry (2026-06-08 catalog refresh, Tier 3). Flagship released 2026-04-23; the successor OpenAI positions above GPT-4o and the o-series.

Updated from Kling 2.0 to Kling 3.0 (released 2026-02-05): native audio, multimodal input, up to 15s clips. (Interim: 2.1, 2.5 Turbo, 2.6.)

Updated to Luma Ray3 (released 2025-09-18), the model behind the Dream Machine app. (Interim: Ray2 early 2025; later Ray3 Modify 2025-12-18, Ray3.14 2026-01-26.)

Updated from v6 to V8.1 (released 2026-04-30) per updates.midjourney.com. (Interim: v7 2025-04-03, v8.0 Alpha 2026-03-17.)

Verified current vs Mistral docs (mistral-large-2512). Corrected release date to 2025-12-02 (Mistral 3 family launch).

Repointed from parakeet-tdt-1.1b to current flagship parakeet-tdt-0.6b-v2 (English, 2025-05-01); added parakeet-tdt-0.6b-v3 (multilingual, 2025-08-14).

Verified current. o3 remains available in the API — OpenAI is retiring it from ChatGPT only (2026-08-26), with no API change. GPT-5.5 is the newer reasoning-capable flagship.

Deprecated — superseded by GPT-5.4 mini; OpenAI API shutdown 2026-10-23.

Verified current vs Mistral docs. Corrected release date to 2025-12-17 (announcement date).

Updated from Pika 2 to Pika 2.2 (released 2025-02-27): Pikaframes, 1080p, up to 10s. (A 'Pika 2.5' is referenced on the site but unconfirmed by a dated source.)

Repointed to active repo OHF-Voice/piper1-gpl (rhasspy/piper archived 2025-10-06; latest release v1.4.2, 2026-04-02).

Replaced fabricated 'Qwen 3.5 122B' entry (no such repo) with the real current open-weight flagship Qwen3.6-35B-A3B (released 2026-04-17). The real Qwen3.5 flagship was Qwen3.5-397B-A17B (2026-02-16).

Deprecated — standalone QwQ reasoning superseded by Qwen3.x thinking mode.

Updated from Gen-3 to Gen-4.5 (released 2025-12-11) per Runway changelog. (Interim: Gen-4 / Gen-4 Turbo 2025-04.)

Updated original Sora to Sora 2 (released 2025-09-30). OpenAI is winding the product down — app closed (2026-04-26), API shutdown slated 2026-09-24; no current OpenAI video successor is generally available.

Updated from Stable Audio 2.5 to Stable Audio 3.0 (released 2026-05-20). Prior release date (2024-04-03) belonged to the 2.0 line.

Updated from v4 to v5.5 (released 2026-03-26) per suno.com/release-notes. (Interim: v4.5 2025-05, v5 2025-09-23.)

Updated to Allegro v1.5 (released 2025-03-18), the latest named model in Udio's official changelog. (Third-party 'v3.5/v4' claims are unverified.)

Updated from Voyage 3 to voyage-4-large (released 2026-01-15). The voyage-3 series is still accessible but Voyage marks it 'not recommended for new implementations'. (Interim: voyage-3.5 2025-05.)

Initial publication.

Initial publication.

Initial publication.

Initial publication. Snapshot reflects flagship model versions GPT-5.4, Claude Sonnet 4.6, and Gemini 3.1 Pro available in May 2026.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication. Pricing snapshot reflects May 2026 flagship and budget tiers.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Marked deprecated after verification that Codestral 25.08 is the current Codestral release.

Added after verification of Mistral's current coding lineup.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Marked deprecated after verification that GPT Image 2 is now OpenAI's current image generation model.

Initial entry.

Initial entry.

Initial entry.

Initial entry. Sighted on Hugging Face (lastModified 2026-05-06).

Initial entry. Sighted on Hugging Face (lastModified 2026-05-06).

Added after verification of Mistral's current frontier coding lineup.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry. Sighted on Hugging Face (lastModified 2026-02-17).

Initial entry.

Initial entry.

Initial entry.

Added after verification of the Gemini 3 family rollout in current Google model docs.

Added after verification of the Gemini 3 family rollout in current Google model docs.

Added after verification of the Gemini 3 family rollout in current Google model docs.

Added after verification of the Gemini 3 family rollout in current Google model docs.

Initial entry.

Marked deprecated after verification that Gemini 3.1 Flash Live Preview is the current Live API successor.

Initial entry.

Initial entry. Sighted on Hugging Face (lastModified 2026-05-07).

Added after verification of the GPT-4.1 family launch and current availability.

Added after verification of the GPT-4.1 family launch and current availability.

Initial entry.

Marked deprecated after OpenAI listed GPT-4.1 as the replacement in its deprecations guide.

Initial entry.

Initial entry.

Added after verification of the current OpenAI speech generation lineup.

Added after verification that GPT Image 2 is the current OpenAI image generation model.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry. Sighted on Hugging Face (lastModified 2025-05-22).

Initial entry. Sighted on Hugging Face (lastModified 2025-05-22).

Initial entry.

Added after verification of Mistral's current reasoning lineup.

Initial entry.

Initial entry.

Added after verification of Mistral's current local-deployment lineup.

Initial entry.

Marked deprecated after verification that Mistral 7B no longer appears in Mistral's current model overview.

Initial entry.

Initial entry. Sighted on Hugging Face under Mistral-Large-3-675B-Instruct-2512 family.

Added after verification of Mistral's current frontier multimodal lineup.

Initial entry.

Marked deprecated after verification that OCR 3 is the current Mistral OCR offering.

Added after verification of Mistral's current frontier multimodal lineup.

Initial entry.

Marked deprecated after verification that Mixtral 8x7B no longer appears in Mistral's current model overview.

Initial entry.

Initial entry.

Initial entry.

Added after verification of Mistral's current OCR lineup.

Initial entry.

Initial entry.

Initial entry.

Marked deprecated after verification that GPT-4o mini TTS is OpenAI's current speech generation model.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry. Sighted on Hugging Face (lastModified 2026-04-24).

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry.

Initial entry. Sighted in Gemini API docs on 2026-05-07.

Initial entry. Sighted in Gemini API docs on 2026-05-07.

Initial entry.

Initial entry.

Backfilled lineage entry; v3 has been the recommended Whisper since 2023.

Initial entry.

Initial entry.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication. Reflects post-August-2025 vendor policy changes (Anthropic opt-in default, retention extensions) and pre-August-2026 EU AI Act enforcement timeline.

Initial publication.

Initial publication.

Initial publication. Snapshot reflects Cursor (Pro/Pro+/Ultra), GitHub Copilot (Pro/Pro+/Business/Enterprise), and Claude Code via Claude Pro/Max plans.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Initial publication.

Updated to recommend Whisper large-v3 over medium-en for English. Throughput numbers updated for new model.

Added Apple Silicon sizing notes; clarified break-even math.

Initial publication.