Devstral
Mistral's 24B Apache 2.0 model built for agentic coding, with All Hands AI.
About
Devstral (May 2025) is Mistral AI's 24B-parameter code model built with All Hands AI (the OpenHands team) specifically for agentic software engineering — exploring codebases, using tools, and completing multi-step tasks — rather than plain autocomplete. It scored 46.8% on SWE-Bench Verified at release (Mistral-reported), the best open result then, and runs on a single RTX 4090 or a 32GB Mac. Apache 2.0, 128K context, also served via Mistral's API. Later followed by Devstral Small 1.1 and Devstral 2, which score higher, but the original remains a strong local agent base.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Mistral-reported: 46.8% on SWE-Bench Verified at release, ahead of all open models then available.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Multiple sources read this session (TestingCatalog, VentureBeat, DeepNewz): May 21 2025, 24B, Apache 2.0, SWE-Bench Verified 46.8% (Mistral-reported), 128K context, All Hands AI collaboration, Mistral API $0.10/$0.30 per 1M.; License clean (Apache 2.0). Devstral 2 (separate catalog entry) noted as successor..