MistralbalancedTool use

Devstral Medium

Devstral Medium is a high-performance code generation and agentic reasoning model developed jointly by Mistral AI and All Hands AI. Positioned as a step up from Devstral Small, it achieves 61.6% on SWE-Bench Verified, placing it ahead of Gemini 2.5 Pro and GPT-4.1 in code-related tasks, at a fraction of the cost. It is designed for generalization across prompt styles and tool use in code agents and frameworks. Devstral Medium is available via API only (not open-weight), and supports enterprise

131K context·~98K words·
Context window131Ktokens

Context window

This model accepts 131K tokens in one request (~98K words of text).

Context window size131K tokens
4K32K128K1M10M

What fits in one request

  • Short document
    About 1,500 words of text
    Fits
  • Long document
    About 37K words of text
    Fits
  • Small codebase
    About 150K words of text
    Won't fit
  • Full novel
    About 375K words of text
    Won't fit

Specifications

Context size, pricing, and release info in one place.

Context window
131,072 tokens (131K)
Speed tier
balanced
Provider
Mistral
Release date
Jul 2025

Capabilities

See which features this model supports, such as vision, tools, and streaming.

Supported (4)
Tool use
Supported
Function calling
Supported
Streaming
Supported
Prompt caching
Supported
Not supported (4)
Vision
Not supported
Extended thinking
Not supported
Web search
Not supported
Batch API
Not supported

Best for

Jump to a guide or ranking that matches each workload.

Compare Devstral Medium

Open a side-by-side comparison with one click.

Frequently asked questions

Short answers about context size and how this model behaves.

Devstral Medium has a context window of 131K tokens (131,072 tokens). This covers most professional use cases including large code files, lengthy reports, and long conversation histories.

More from Mistral

Other models by Mistral in our catalog.

Powered by Mem0

Use a smaller model.
Get better results.

Mem0 gives your AI long-term memory so you stop re-sending context on every call. That means you can use a smaller, faster, cheaper model — and still get better answers.

Example: a multi-turn chat session

Without Mem0~128K tokens sent
Full history
Repeated info
Old context
With Mem0~20K tokens sent
Key memories
Current turn

80% less to send — works with any model