Mistral AIProvisional

Devstral 2

Devstral 2 / Default · devstral-latest

Mistral coding-agent model for software engineering tasks.

Objective score

31.2

Confidence

86%

Price evidence

Pricing evidence pending

Support this model Compare trend Official source

Scenario evidence

Six task-specific rankings

Feature development#42Obj. 31.20 · 0%Debug and repair#44Obj. 42.40 · 0%Refactoring#47Obj. 38.70 · 0%Code review#50Obj. 37.10 · 0%Large-repo agent work#40Obj. 28.80 · 0%Frontend / UI#48Obj. 37.20 · 0%

Evidence state

Signal coverage

Benchmarks: 3
Rank points: 7
Usage: 2
Agent: 0
Sources: 5

Objective evidence

Benchmarks and public signals

Swebench: 53.8
Livecodebench: —
Aider: —
Artificial Analysis: 31.3
Openrouter Usage: 0
Kilo Usage: —

Persistent history

Recent ranking points

Aug 1#42 · 28

Aug 1#42 · 28

Aug 1#42 · 28

Jul 31#42 · 28

Jul 31#42 · 28

Jul 31#42 · 28

Usage signals

OpenRouter0%

Overall0%

Agent and tool signals

Evidence is being collected. Missing data is not treated as zero.

Source trail

Official Mistral docs list Devstral 2 as a frontier code agents model.SWE-bench Verified · ok Artificial Analysis Coding Agents · ok WebDev Arena · ok Kilo leaderboard · ok Mistral model overview · ok