Model · Alibaba · Text
Qwen3.8 27B
Alibaba's current open-weight workhorse — a 27B dense model under Apache 2.0, natively multimodal, and the model most commonly run on a single 24GB consumer GPU. Supersedes the Qwen3.6 generation.
- Modality
- Text
- License
- Apache 2.0 (Open)
- Parameter size
- 27B (dense)
- Context window
- 262,144 tokens · up to 1M 262,144 tokens native, extensible to roughly 1M with YaRN scaling. Static YaRN scaling can hurt short-text performance, so enable it only for genuinely long inputs.
- Released
- August 1, 2026
- Last verified
- October 28, 2026
- Runs locally
- Yes
- Also handles
- vision, reasoning, code
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 32 GB (48 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 24 GB (32 GB recommended) of video memory.
- No graphics card
- Not practical. Too slow without a graphics card or Apple Silicon.
Fits a 24 GB graphics card or a 32 GB Mac in Ollama, LM Studio or MLX (the MLX 4-bit build is the most downloaded community build).
Numbers are for: 4-bit build of the 27B (about 17 GB). Our estimate from the model’s size — the maker publishes no figure.
Strengths
- Apache 2.0 — no commercial restrictions
- Dense 27B fits a single 24GB consumer GPU, so it self-hosts without a cluster
- Native vision-language across images and video
- Strong agentic coding: SWE-bench Pro 61.7%, Terminal Bench 2.1 73.0%
- Thinking is controllable through a reasoning_effort parameter
Weaknesses
- Thinking mode is on by default, which adds latency unless you turn it down
- Video preprocessing is configured conservatively for inference efficiency
- Alibaba publishes only a month-level release date (August 2026), not a day
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | Apache 2.0; FP8 quantization also published |
| Alibaba Cloud Model Studio | hosted-api | API key required |
Used in solutions
Official sources
- Model card model
- Qwen blog announcement
Change log
- — 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.
- — 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.
Esc