Devstral Small 2 24B Instruct 2512
Mistral AI
Code Multilingual Tool Calls
Devstral Small 2 24B Instruct 2512 is a 24.01-billion-parameter dense transformer from Mistral AI, optimized for agentic software engineering and multi-file code editing on consumer hardware. It scores 68.0% on SWE-bench Verified, rivaling models up to five times its size. The model supports tool calling and 11 languages including English, French, German, and Chinese. With a 384K context window and flash attention, it runs on a single GPU and can be self-hosted with GGUF quantization.
Hardware Configuration
Optional — for precise deployment recommendations
| Quantization | Quality | Size | Fit |
|---|---|---|---|
| Q8_0 | High | 23.33 GB | — |
| Q8_K_XL | High | 27 GB | — |
| Q6_K | High | 18.02 GB | — |
| Q6_K_XL | High | 19.36 GB | — |
| Q5_K_M | Medium | 15.61 GB | — |
| Q5_K_S | Medium | 15.19 GB | — |
| Q5_K_XL | Medium | 15.61 GB | — |
| Q4_K_M | Medium | 13.35 GB | — |
| Q4_K_S | Medium | 12.62 GB | — |
| Q4_K_XL | Medium | 13.51 GB | — |
| Q4_0 | Medium | 12.57 GB | — |
| Q4_1 | Medium | 13.85 GB | — |
| Q3_K_M | Low | 10.69 GB | — |
| Q3_K_S | Low | 9.69 GB | — |
| Q3_K_XL | Low | 11.04 GB | — |
| Q2_K | Low | 8.28 GB | — |
| Q2_K_L | Low | 8.43 GB | — |
| Q2_K_XL | Low | 8.65 GB | — |
Last updated: March 6, 2026