Mistral Medium 3.5 is a 128B dense language model that unifies instruction-following, reasoning, and coding in a single set of weights. It features a 256k context window and configurable reasoning effort, making it suitable for complex, long-running tasks. The model achieves 77.6% on SWE-Bench Verified and is available as open weights on Hugging Face, empowering engineers to run self-hosted inference.
Mistral Medium 3.5
A 128B model for coding, reasoning, and long tasks
Mistral Medium 3.5 介绍
核心功能
- 128B dense model with state-of-the-art coding and reasoning performance
- 256k token context window for handling long documents and conversations
- Configurable reasoning effort per request for flexible responses
- Open weight release under modified MIT license for self-hosting
- Scores 77.6% on SWE-Bench Verified and 91.4 on τ³-Telecom
典型使用场景
- Software engineers using Vibe CLI for asynchronous, cloud-based coding agents
- Developers needing a powerful LLM for complex multi-step reasoning tasks
- Teams running self-hosted inference for data-sensitive applications
- Users of Le Chat's Work mode for research, inbox triage, and cross-tool workflows
为什么适合初创团队
Mistral Medium 3.5 offers startup-friendly open weights, enabling self-hosting to avoid per-token costs and maintain data privacy. Its strong performance on coding and reasoning tasks accelerates development cycles, while the 256k context supports long-form agentic workflows. Startups can leverage the same model for both quick queries and complex, multi-step tasks without switching between different models.
可替代选择
GPT-4, Claude 3.5, Gemini 2.0, Qwen 3.5, Llama 3.1
常见问题
What is the context length of Mistral Medium 3.5?
It supports a 256k token context window.
Is the model available as open source?
Yes, it is released with open weights on Hugging Face under a modified MIT license.
How does the pricing work for the API?
The API costs $1.5 per million input tokens and $7.5 per million output tokens.
Can I run Mistral Medium 3.5 on my own hardware?
Yes, it can be self-hosted on as few as four GPUs, with deployment guides available.
