Model · Mistral AI · Text
Mistral Medium 3.5
Mistral's frontier-class multimodal model for agentic and coding use cases. It sits between the largest flagship tier and the lighter Small line while remaining open-weight.
- Modality
- Text
- License
- Modified MIT (Open weights)
- Context window
- 256,000 tokens
- Released
- April 28, 2026
- Last verified
- October 28, 2026
- Runs locally
- Yes
- Also handles
- vision, code
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 96 GB (128 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- Server only. Needs a server with several graphics cards, about 96 GB (160 GB recommended) of video memory in total — not a desktop PC.
- No graphics card
- Not practical. Too slow without a graphics card or Apple Silicon.
Needs a 96 GB or 128 GB Mac (MLX and Ollama builds exist) and generates text slowly because all 128B parameters are used for every word.
Numbers are for: 4-bit build of the dense 128B model (about 78 GB). Our estimate from the model’s size — the maker publishes no figure.
Strengths
- 256k context window
- Strong fit for agentic workflows and code-heavy tasks
- Open weights under a Modified MIT license
Weaknesses
- Heavier deployment footprint than smaller open models
- Still a newer model line with less third-party deployment guidance
Try it
| Where | Type | Notes |
|---|---|---|
| Mistral La Plateforme | hosted-api | API key required |
| Mistral model card | weights | Modified MIT / open weights |
Official sources
- Model card docs
- Announcement announcement
Change log
- — 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.
- — 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.
- — Added after verification of Mistral's current frontier multimodal lineup.
Esc