CITATION — REFERENCE ENTRY
Devstral — Mistral AI News
- Key
- mistral2025devstral
- Authors
- Mistral AI
- Issued
- 2025-5-21
- Type
- webpage
- Container
- Mistral AI News
Raw CSL JSON
{
"URL": "https://mistral.ai/news/devstral/",
"note": "Model announcement by one of the two collaborating parties; benchmark figures are self-reported.",
"type": "webpage",
"title": "Devstral",
"author": [
{
"literal": "Mistral AI"
}
],
"issued": {
"date-parts": [
[
2025,
5,
21
]
]
},
"accessed": {
"date-parts": [
[
2026,
8,
29
]
]
},
"container-title": "Mistral AI News"
}
Claims
-
Devstral is an agentic language model for software engineering built as a collaboration between Mistral AI and All Hands AI and released under the Apache 2.0 licence. Mistral reports it scoring 46.8% on SWE-Bench Verified when evaluated inside the OpenHands scaffold, exceeding much larger models such as DeepSeek-V3-0324 (671 billion parameters) and Qwen3 232B-A22B under the same scaffold. Mistral states the model is light enough to run on a single RTX 4090 or a Mac with 32 GB of RAM, and describes it as a research preview.
"Devstral achieves a score of 46.8% on SWE-Bench Verified ... light enough to run on a single RTX 4090 or a Mac with 32GB RAM ... Devstral is a research preview"
Available in