Devstral 24B
by Mistral AI · mistral family
24B
parameters
text-generation code-generation reasoning
Devstral 24B is Mistral's dedicated coding agent model, fine-tuned from Mistral Small 3.1 for software engineering tasks. It excels at code generation, repository-scale understanding, debugging, and agentic coding workflows. Ranked #1 among open-source coding agent models at launch. At Q4 it fits on 24 GB GPUs — ideal for developers who want a local alternative to cloud-based coding assistants like GitHub Copilot.
Quick Start with Ollama
ollama run 24b-q4_K_M | Creator | Mistral AI |
| Parameters | 24B |
| Architecture | transformer-decoder |
| Context | 128K tokens |
| Released | May 21, 2025 |
| License | Apache 2.0 |
| Ollama | devstral |
Quantization Options
| Format | File Size | VRAM Required | Quality | Ollama Tag |
|---|---|---|---|---|
| Q4_K_M rec | 14 GB | 17 GB | | 24b-q4_K_M |
| Q8_0 | 25 GB | 29 GB | | 24b-q8_0 |
| F16 | 48 GB | 53 GB | | 24b-fp16 |
Compatible Hardware
Q4_K_M requires 17 GB VRAM
Benchmark Scores
72.0
mmlu