AI/TLDR

Devstral Medium & Small 1.1 (25.07)

Mistral's July 2025 Devstral refresh — an open Apache-2.0 Small 1.1 plus an API-only Devstral Medium for agentic coding.

Overview

Devstral Medium and Devstral Small 1.1, released July 10, 2025, were Mistral AI's mid-2025 update to its Devstral coding-agent line, developed in collaboration with All Hands AI. Devstral Small 1.1 is an open Apache 2.0 model with 24 billion parameters, while Devstral Medium is a higher-performing model available only through the Mistral API and enterprise deployment.

Devstral Small 1.1 scores 53.6% on SWE-Bench Verified — an improvement of about 6.8 points over the original Devstral Small — and was positioned as state-of-the-art among open models for code agents. Devstral Medium reaches 61.6% on SWE-Bench Verified, which Mistral compared favourably to Gemini 2.5 Pro and GPT-4.1 at roughly a quarter of the price. Both models use a 128K-token context window.

Devstral Small 1.1 is fine-tuned from Mistral-Small-3.1-24B-Base-2503 and is text-only (the vision encoder was removed for coding-agent use). It supports both Mistral's function-calling format and XML tool-call formats, and emphasises generalization across prompts and agentic frameworks. Both models in this generation have since been deprecated in favour of the December 2025 Devstral 2 family.

Released2025-07-10
LicenseDevstral Small 1.1: Apache 2.0; Devstral Medium: proprietary (API only)
WeightsOpen weights
ParametersDevstral Small 1.1: 24B; Devstral Medium: not disclosed
Context128K
ArchitectureDense decoder-only Transformer; Devstral Small 1.1 fine-tuned from Mistral-Small-3.1-24B-Base-2503 (text-only, vision encoder removed)
Knowledge cutoff2025
ModalitiesText
StatusDeprecated

Benchmarks

  1. SWE-Bench Verified (Devstral Small 1.1)53.6%
  2. SWE-Bench Verified (Devstral Medium)61.6%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

Pricing

Input$0.10 / 1M tokens (Devstral Small 1.1); $0.40 / 1M tokens (Devstral Medium) per 1M tokens
Output$0.30 / 1M tokens (Devstral Small 1.1); $2.00 / 1M tokens (Devstral Medium) per 1M tokens

Mistral API pricing as published in the July 2025 announcement.

Pricing source ↗

Strengths

  • Devstral Small 1.1 is open Apache 2.0 and was state-of-the-art among open code-agent models at release
  • Devstral Medium reaches 61.6% SWE-Bench Verified at a fraction of comparable proprietary-model pricing
  • 128K-token context for multi-file and long-prompt software-engineering workflows
  • Supports both Mistral function-calling and XML tool-call formats
  • Built with All Hands AI for strong generalization across agentic frameworks

Best for

  • Open, self-hostable coding agents (Devstral Small 1.1)
  • Higher-accuracy agentic software engineering via the Mistral API (Devstral Medium)
  • Resolving real GitHub issues in OpenHands-style scaffolds
  • Enterprise on-premise coding deployments and custom fine-tuning (Devstral Medium)

How to access

ProviderModel ID
Mistral AI ↗devstral-medium-2507
Mistral AI ↗devstral-small-2507

Devstral — every version

The full lineage of the Devstral line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.

VersionReleasedContextLicense
Devstral 2 / Devstral Small 2current2025-12-09256KDevstral 2: Modified MIT; Devstral Small 2: Apache 2.0
Devstral Medium & Small 1.1 (25.07)2025-07-10128KDevstral Small 1.1: Apache 2.0; Devstral Medium: proprietary (API only)
Devstral Small (25.05)2025-05-21128KApache-2.0

FAQ

How do Devstral Medium and Devstral Small 1.1 differ?

Devstral Small 1.1 is a 24B open Apache 2.0 model scoring 53.6% on SWE-Bench Verified, suitable for self-hosting. Devstral Medium is API-only (proprietary) and scores 61.6% on SWE-Bench Verified, with higher accuracy and enterprise deployment options. Both use a 128K context window and were released July 10, 2025.

Is Devstral Small 1.1 open source?

Yes. Devstral Small 1.1 (model ID devstral-small-2507) is released under the permissive Apache 2.0 license with open weights on Hugging Face, fine-tuned from Mistral-Small-3.1-24B-Base-2503. Devstral Medium, by contrast, is proprietary and available only through the Mistral API or enterprise agreements.

What did Mistral compare Devstral Medium to?

Mistral stated that Devstral Medium's 61.6% on SWE-Bench Verified surpasses Gemini 2.5 Pro and GPT-4.1 for roughly a quarter of the price, positioning it on a new cost/performance point for agentic coding.

Are these models still available?

Both Devstral Medium and Devstral Small 1.1 have been deprecated and superseded by the December 2025 Devstral 2 family (Devstral 2 and Devstral Small 2). The open Devstral Small 1.1 weights remain downloadable on Hugging Face.