mistralai/Mistral-Medium-3.5-128B · Hugging Face
A dense 128B model with 256K context and optional reasoning mode for agentic tasks.
Mistral AI has launched Mistral Medium 3.5, a flagship 128B parameter dense model with a 256K context window. This unified model handles instruction-following, reasoning, and coding tasks, replacing its predecessor Mistral Medium 3.1 and Magistral in Le Chat, as well as Devstral 2 in the Vibe coding agent. Key features include multimodal input (text and images), configurable reasoning effort per request—allowing the same model to toggle between fast instant replies and deep reasoning for complex agentic runs. The vision encoder was trained from scratch to handle variable image sizes and aspect ratios.
Mistral Medium 3.5 offers native function calling, JSON output, strong system prompt adherence, and multilingual support for dozens of languages. It is released under a Modified MIT License, permitting both commercial and non-commercial use, with exceptions for high-revenue companies. Recommended settings include using reasoning_effort="high" for complex tasks and agentic coding, with temperature at 0.7. For reasoning_effort="none", temperature can range from 0.0 to 0.7 depending on the task. This release marks a significant step in unifying advanced AI capabilities in a single, open-source model.
- Dense 128B parameter model with 256K context window, supporting multimodal input (text and images)
- Configurable reasoning effort per request, enabling fast replies or deep reasoning for agentic tasks
- Released under Modified MIT License for commercial and non-commercial use, with exceptions for high-revenue companies
Why It Matters
Mistral's unified 128B model with configurable reasoning and 256K context sets a new standard for open-source AI.