Autor: Jakub Rusinowski · Ostatnia aktualizacja: 15 sierpnia 2026
Druga generacja specjalisty od kodowania Mistral AI, wydana w kwietniu 2026. Wariant 123B Sparse osiąga 71,6% na SWE-bench Verified — w chwili premiery najwyższy wynik open-source w zadaniach agentów programistycznych. Zbudowany do agentowej inżynierii oprogramowania: edycji wielu plików, nawigacji po repozytorium i programowania sterowanego testami. Licencja Apache 2.0.
| Licence | What it permits | Applies to |
|---|---|---|
Apache-2.0 | Commercial use permitted Commercial use permitted. No usage restrictions beyond attribution. | Devstral-2 123B, Devstral-2 22B, Devstral Small 24B |
| Devstral-2 123B | Min 75 GB VRAM · Q4_K_M · 262,144 ctx · ollama run devstral:123b |
| Devstral-2 22B | Min 14 GB VRAM · Q4_K_M · 128,000 ctx · ollama run devstral:22b |
| Devstral Small 24B | Min 15 GB VRAM · Q4_K_M · 128,000 ctx · ollama run devstral:24b |
The cheapest GPU that runs Devstral locally (min 14 GB VRAM) is the AMD Radeon RX 9060 XT 16GB (16 GB).
Install Ollama then run: ollama run devstral:123b
Minimum VRAM: 14 GB. For best results use Q4_K_M quantization.
Devstral needs about 14 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Devstral-2 123B (75 GB, Q4_K_M); Devstral-2 22B (14 GB, Q4_K_M); Devstral Small 24B (15 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Yes — Devstral runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
Q4_K_M is the best balance of quality and VRAM for Devstral in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run devstral:123b. This downloads Devstral and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.