Back to Model Tracker
Mistral Active

Devstral 2

Mistral’s frontier code-agents model, built for software engineering tasks, tool use, codebase exploration, and multi-file edits.

Devstral Released 2025-12-09 Updated May 20, 2026
Post Reboot Practical Scores

Capability profile

83.4
Reasoning 82.0
Coding 92.0
Writing 76.0
Speed 87.0
Cost efficiency 88.0
Image understanding 20.0
Pricing
Input
$0.40 / 1M tokens
Output
$2.00 / 1M tokens
Cached input
See Mistral pricing
Context
256K tokens
Free tier
No / unknown
API
Available
Subscription
Mistral API / La Plateforme
Open-source status
Closed
Best Use Cases
Best for
  • Code agents
  • repository exploration
  • multi-file edits
  • and tool-heavy software engineering workflows.
Best workflows
  • Agentic coding assistants
  • repo analysis
  • structured tool calling
  • and automated engineering tasks.
Weakest at
  • Broad premium generalist reasoning outside code-centric work.
Not ideal for
  • Teams that want one model to cover every non-coding workflow equally well.
Post Reboot Take

Devstral 2 matters because it is aimed directly at the coding-agent stack instead of trying to be another broad general-purpose flagship.

Mistral’s frontier code-agents model, built for software engineering tasks, tool use, codebase exploration, and multi-file edits.

Post Reboot take

Devstral 2 matters because it is aimed directly at the coding-agent stack instead of trying to be another broad general-purpose flagship.

Strengths

Best for: Code agents, repository exploration, multi-file edits, and tool-heavy software engineering workflows.

Weakest at: Broad premium generalist reasoning outside code-centric work.

Best workflows: Agentic coding assistants, repo analysis, structured tool calling, and automated engineering tasks.

Example use cases: Bug-fix agents, codebase search and patch flows, implementation assistants, and CI-connected coding tools.

Sources used for this profile