Mistral Devstral Small 2
24B Apache-2.0 agentic coder that runs on a single RTX 4090 — 68% on SWE-bench Verified.
About
Devstral Small 2 (December 9, 2025) is Mistral's laptop-friendly agentic coding model: 24B parameters, 256K context, fine-tuned specifically for software-engineering agent tasks — exploring codebases, editing multiple files in one pass, reading diffs, and calling tools in agentic loops — rather than generating boilerplate from scratch. It scores 68.0% on SWE-bench Verified (provider-reported), competitive with models up to five times its size, and runs on a single GPU or even CPU-only setups. The licensing is the reason to care: pure Apache 2.0, unlike its 123B sibling Devstral 2, which ships under a modified MIT license restricting organizations above $20M/month revenue. Available via Mistral's API (reported $0.10/$0.30 per 1M input/output tokens post-free-period), Ollama as devstral-small-2, and Hugging Face. It even takes image inputs for multimodal agent use.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- SWE-bench Verified 68.0% — provider-reported
- Up to 7x cost-efficiency vs Claude Sonnet on real-world tasks — provider-reported
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: Dec 9, 2025 release verified across TechCrunch, Augmenter, Open Source For You, and AIFoss review; 24B, 256K context, Apache 2.0 (vs 123B sibling's modified MIT with revenue clause) — license distinction well documented; SWE-bench 68.0% and $0.10/$0.30 pricing provider-reported — labeled as such; independent review corroborates the positioning.