Sulat.com
AI models
$10 off the fastest DeepSeek V4.1 Flash, Kimi K3 and GLM 5.3 from Synthetic
Merge Gateway logo

Model details

Devstral Small

Devstral Small is an open-weights agentic language model purpose-built for software engineering work, developed jointly by Mistral AI and All Hands AI. It is fine-tuned from Mistral-Small-3.1, with the vision encoder removed before fine-tuning so that the resulting model operates purely on text. This lineage keeps the model focused on code reasoning, tool use, and multi-file editing rather than multimodal perception, and it preserves a long context window inherited from its base, which makes it well suited to navigating and modifying large codebases in a single pass.

At a compact scale of around 24 billion parameters, Devstral Small is light enough to run locally or on modest hardware, while still delivering strong results on the SWE-bench software engineering benchmark, where it is positioned as the leading open-source model. The release improves on the earlier Devstral Small 1.0 with better generalization across prompts and coding environments, and it works naturally with OpenHands and Mistral's function-calling format. Practically, this makes it a good fit for developers who want a self-hosted coding agent that can explore repositories, chain tool calls, and act as the reasoning core of an automated software engineering workflow.

Merge Gatewaymistral/devstral-small-2507devstraldeprecated

Quick Info

Powered by
Provider
Merge Gateway
Model key
mistral/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

Transparent token rates

Compare Devstral Small pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Devstral Small

Merge Gateway

Coverage

Mistral AI's July 10, 2025 announcement introduces Devstral Medium alongside an upgrade to Devstral Small, releasing the new model under the Apache 2.0 license in collaboration with All Hands AI. The page explicitly names the API slug "devstral-small-2507," which corresponds to the Devstral Small 1.1 subject served via Merge Gateway, and confirms its availability at $0.1/M input tokens and $0.3/M output tokens (same pricing as Mistral Small 3.1). Architecture remains at 24B parameters with significant capability improvements over its predecessor. Devstral Small 1.1 achieves 53.6% on SWE-Bench Verified, setting a new state of the art for open models without test-time scaling, and generalizes well across prompts, coding environments, and agentic scaffolds including OpenHands. The model supports both Mistral function calling and XML tool formats, making it adaptable to a wide range of agentic applications. The excerpt is truncated before describing Devstral Medium's 61.6% SWE-Bench Verified score and availability details, so those specifics are not summarized here.

Videos about Devstral Small

More models around Devstral Small