Sulat.com
AI models
Mistral logo

Model details

Devstral Small

Devstral Small sits in Mistral's coding-focused family as a compact, open-weight language model purpose-built for software engineering agents. According to deployment guidance, it is built on a Mistral Small 3.1 24B base and further fine-tuned on coding and tool-use data, giving it a strong starting point for instruction following while keeping the parameter count light enough to run on a single high-end GPU. The model is positioned around SWE-bench style agentic tasks, where the goal is not just to generate code but to explore repositories, edit multiple files, and coordinate tool calls to complete real engineering work.

In practical terms, Devstral Small is meant for developers who want to host or self-deploy a capable coding assistant without paying per-seat SaaS fees, integrating with IDE plugins and agent frameworks such as Continue, Aider, Cline, Claude Code, OpenCode, Hermes Agent, and OpenClaw. Distribution through Ollama and Docker makes the 24B Instruct variant straightforward to pull and run, and the large context window supports whole-repository reasoning across long file trees. Its strengths are agentic exploration, multi-file editing, and tool calling rather than general-purpose chat, making it a sensible fit for teams building internal coding agents or experimenting with local-first software engineering workflows.

Mistraldevstral-small-2507devstraldeprecated

Quick Info

Powered by
Provider
Mistral
Model key
devstral-small-2507
Release date
Jul 10, 2025
Last updated
Jul 10, 2025
Knowledge cutoff
2025-05
Input modalities
Output modalities
Capabilities

Cost

Input token cost
$0.10
Output token cost
$0.30

Limits

Output tokens
128,000 tokens
Context window
128,000 tokens

Latest news about Devstral Small

Mistral

Coverage

Spheron's May 2026 deployment guide covers self-hosting Devstral on GPU Cloud using vLLM, positioning it as a 24B coding-specialized model from Mistral AI trained on SWE-bench agentic tasks. The architecture is described as Mistral Small 3.1 24B base with fine-tuning on coding and tool-use data, with a 128K context win The guide frames Devstral as a cost-effective alternative to per-seat SaaS coding assistants like Cursor Pro ($20/seat) and GitHub Copilot Business ($19/seat), detailing vLLM setup on Spheron, IDE plugin configuration for Continue, Aider, and Cline, and team-size cost math. It also notes that Mistral has since moved it

Mistral

CoverageBenchmark

OpenRouter's listing for mistralai/devstral-small corroborates the official Devstral Small 1.1 (2507) specification: a 24B-parameter open-weight model finetuned from Mistral Small 3.1, released July 10, 2025, under Apache 2.0, with a 128k-token (131,072) context window, a Tekken tokenizer with 131k vocabulary, and a Ma The page also documents deployment options, listing vLLM, Transformers, Ollama, LM Studio, and other OpenAI-compatible runtimes as supported serving backends, and confirms native support for both Mistral-style function calling and XML tool output for integration with agentic scaffolds such as OpenHands and Cline. As a

Videos about Devstral Small

More models around Devstral Small