Devstral Small 2

Devstral Small 2

Mistral · Released Dec 9, 2025
Intelligence #528 / 715
26.4 our score
Speed #79 / 321
140.7 tok/s
Input Price
Not priced
Output Price
Not priced
Context
Not reported

Analysis Summary

Devstral Small 2 is a Mistral model focused on software development workloads. Its coding results are materially stronger than its general reasoning results, and terminal performance is better than the other Mistral variants evaluated here. Agentic results remain modest, so it should not be treated as a fully autonomous engineering system.

The model fits code completion, small repository changes, test generation, and developer-assistance tasks where a human reviews the result. Its instruction-following and long-context measures are limited, making complex implementation plans, broad refactors, and customer-facing technical explanations less dependable. Pricing data is not supplied, so its value case rests on its compact positioning rather than a verified cost advantage.

Adopt it as a coding specialist for bounded tasks and internal workflows. Keep stronger reasoning models available for architecture, review, and multi-step tool use.

Assessed August 9, 2026

Editorial notes

Devstral Small 2 offers Mistral's strongest coding profile in this group, with better terminal performance than its siblings and low-cost positioning, but reasoning and agentic reliability remain limited.

Rankings consider pricing, capabilities, benchmarks, and real-world applicability and are refreshed as new models launch. Feedback?

DFO Verdict

Devstral Small 2 offers Mistral's strongest coding profile in this group, with better terminal performance than its siblings and low-cost positioning, but reasoning and agentic reliability remain limited.

#528 of 715 overall Down 12 this week

Benchmark scores

GPQA Diamond 53.2%
HLE 3.4%
MMLU Pro 67.8%
AIME 2025 34.3%
SciCode 28.8%
LiveCodeBench 34.8%
TerminalBench Hard 16.7%
τ²-Bench 23.4%
IFBench 31.2%
LCR 24%

Magenta = intelligence · Ink = technical/agentic · Cyan = content & long-context · Grey = community benchmarks. Data: Artificial Analysis, Hugging Face.

17.7 Intelligence Index·29.3 Coding Index·10.8 Agentic Index·34.3 Math Index

How Devstral Small 2 compares

Devstral Small 2 ranks #196 of 428 AI models we track for overall intelligence, #113 of 201 for coding, #116 of 183 for agentic tasks. Devstral Small 2 is currently free to use via OpenRouter.

Position in the field
Intelligence: smarter than 26% of models #528
Speed: faster than 75% of models #79
worst in fieldmedianbest in field
Price vs frontier peers · $ per 1M tokens
Devstral Small 2 $0.00 in $0.00 out
OpenAI: GPT-5.6 Sol $2.00 in $10.00 out
Claude Opus 5 $5.00 in $25.00 out
SpaceXAI: Grok 4.6 $2.00 in $6.00 out

Dark bar = input · light bar = output, scaled to the priciest peer.

Context window vs peers · tokens
Devstral Small 2 0

1M tokens ≈ 8 full-length novels or ~2,500 pages of business documents in a single request.

Intelligence3Technical2.6Value0Content4.3
Performance profile

Strongest on business fit. The pulled-in value corner is the trade-off, and if the shape matters more than the price, this is your model.

Compare shapes side-by-side →

Frequently asked questions about Devstral Small 2

How much does Devstral Small 2 cost?

Devstral Small 2 is currently available for free via OpenRouter.

Is Devstral Small 2 good for coding?

On our coding benchmark index, Devstral Small 2 ranks #113 of 201 models, placing it in the broader range of the field for code generation and debugging.

Who created Devstral Small 2?

Devstral Small 2 is developed by Mistral and was released on December 9, 2025.

© 2026 Design for Online Ltd. Registered in England and Wales No. 10328553. VAT Registered. Design for Online® and Forerunner® are registered trademarks.