Mistral
Mistral's Devstral Medium is a coding-focused model with modest benchmark performance across the board, suitable for lighter development ...
Model details
Devstral Medium is positioned as a coding-focused language model intended for software engineering workflows such as code generation, repository navigation, and agent-style tool use. Third-party coverage describes it as a text-in, text-out model with tool calling and temperature control, released in July 2025 with a knowledge cutoff in May 2025, and distributed as open weights so developers can self-host and integrate it into local pipelines. Its large context and output window make it suitable for handling sizable codebases, multi-file refactors, and long generation passes in a single request, which is particularly useful for agentic assistants that need to reason across many files at once.
On standard coding evaluations the model shows a mixed profile: one comparison source reports it leading OpenAI's o1-preview on SWE-Bench Verified while being dramatically cheaper per token, whereas another independent reviewer characterizes its benchmark results as modest, so prospective users should validate performance on their own codebases rather than rely on headline scores. Pricing is clearly documented at $0.40 per million input tokens and $2.00 per million output tokens, and it carries an identical 128,000-token context and maximum output ceiling, giving generous headroom for tool-augmented coding sessions. Practical fit centers on teams that want an open-weights coding model with strong tool-calling support and a budget-friendly inference cost, especially for IDE plugins, automated refactoring, and CI-driven code review pipelines.
Mistral
Mistral's Devstral Medium is a coding-focused model with modest benchmark performance across the board, suitable for lighter development ...
Mistral
Artificial Analysis flags Devstral Medium as a deprecated model, noting that Mistral has launched a newer model, Devstral 2, which it recommends considering instead. The page states that only default 10k input token workload benchmarking continues, while results for other workloads remain historical and are no longer u The page's provider comparison sections for fastest output speed, lowest latency, and lowest blended price all show zero providers and no data available, confirming that live benchmarking data for Devstral Medium is no longer maintained. Artificial Analysis directs readers to the dedicated models page for Devstral Medi
Mistral
Compare Devstral Medium vs GPT-5.1 Codex Mini: input $0.4/M vs $0.25/M, output $2/M vs $2/M tokens. GPT-5.1 Codex Mini is 7% cheaper overall. Full API cost breakdown, context window, and benchmark comparison.