Skip to main content

Decision brief

Mistral Vibe Review: Low-Cost European Agent or Claude Code Stand-In?

Who Mistral Vibe fits, where it trails Claude Code and Codex, and the plan, data, and approval settings to decide before a team uses it.

Reading time
3 min read
Sources
21 cited

GILD verdict

Choose Mistral Vibe when a team wants an open-source terminal agent on a European provider's models at a flat monthly price, for changes an engineer scopes and watches.

Mistral documents Pro at US$14.99 a month with all-day coding in the CLI, IDE, or web, and the CLI is Apache 2.0, so a trial costs little. You take on review of every change and quota planning, and you must switch off the data settings Mistral leaves on by default. Move to Claude Code or Codex when the work is long, multi-step change across a large codebase that nobody watches turn by turn.

Mistral Vibe

AI Coding Agent by Mistral AI

+18 Mixed
Product page

Fit and limits

Who it fits

  • The team wants the agent's source open for audit: the CLI is Apache 2.0, written in Python, with its system and tool prompts kept as Markdown files in the repository.
  • Model traffic should go to a European company. Mistral Medium 3.5 is Vibe's built-in model, and Enterprise can deploy Vibe on-premises, in a private cloud, or on Mistral Cloud with full data residency.
  • Engineers want one agent across the terminal and the editor. Vibe runs as the CLI and a VS Code extension, in JetBrains IDEs and Zed through ACP, and in remote Vibe Code Web sessions that can open pull requests.
  • Some work must run offline: the CLI talks to any OpenAI-compatible server, and Mistral documents vLLM, llama.cpp, LM Studio, and Ollama, with Devstral Small 2 for a single 24 GB GPU.

Where it falls short

  • Plan usage is a monthly budget shared across Vibe Code, Studio, and the API; when it is spent, Vibe Code stops until the next period unless pay-as-you-go is on.
  • Defaults favor speed over control: the default agent applies file edits without asking, telemetry is on, and input and output on Vibe plans other than Enterprise train Mistral's models until a user opts out.
  • Windows works but is not officially supported; the README says Mistral targets UNIX environments.
  • Vibe 3.0 removed the built-in plan agent, so a read-only planning step needs a custom agent with the edit and write tools disabled.

Implementation reality

  1. Buy the plan before the key

    Pro at US$14.99 per month and Team at US$24.99 per user per month, both before tax, include usage shared across Vibe Code, Studio, and the API. The CLI signs in with a Mistral account or takes an API key. Decide whether pay-as-you-go may continue past the included usage, and who watches the balance.

  2. Set the approval mode

    Vibe has four approval modes: ask, accept-edits (the default), smart-approve, and auto-approve. Mistral describes smart-approve as a classifier with no sandbox, and reserves auto-approve for containers, CI runners, and disposable VMs. An administrator can set enable_auto_approve to false, which hides auto-approve and, from CLI 3.0.1, ignores --yolo.

  3. Push the config from the admin panel

    Admin config sets models, providers, allowed models, MCP servers, connectors, tool permissions, and telemetry for the CLI and the VS Code extension, per organization or workspace. It overrides local files and flags. It needs CLI 2.24.2 and VS Code extension 1.16 or later.

  4. Settle data handling

    Opt the organization out of training, or buy Enterprise, which is opted out by default. Zero data retention covers Vibe CLI calls on paid plans once Mistral approves a request. Regional inference on api.eu.mistral.ai is documented for the API at 1.1 times list price and does not make the control plane regional.

  5. Track the release pace

    Mistral shipped CLI 2.26.0 on Oct 6, 2026, then 3.0.0 on Oct 19 and 3.0.2 on Oct 22. Pin a version for a pilot and read the changelog before each upgrade, since 3.0 removed the plan agent.

What to measure in your pilot

  • How accepted, reviewed changes per month on Pro compare with Claude Code and Codex on the same repositories.
  • Whether Vibe CLI traffic can be pinned to EU processing the way the API's regional endpoint is.
  • How much of a heavy user's month the Pro budget covers on each hosted model.

Sources

Official documentation
21
  1. Mistral VibeOfficial documentationmistral.ai
  2. Mistral pricingOfficial documentationmistral.ai
  3. Mistral: Introducing Devstral 2 and Mistral Vibe CLI (Dec 9, 2025)Official documentationmistral.ai
  4. Mistral: Mistral Vibe 2.0 (Jan 27, 2026)Official documentationmistral.ai
  5. Mistral: Remote agents in Vibe, powered by Mistral Medium 3.5 (May 22, 2026)Official documentationmistral.ai
  6. Mistral docs: Vibe CodeOfficial documentationdocs.mistral.ai
  7. Mistral docs: Install and setupOfficial documentationdocs.mistral.ai
  8. Mistral docs: API keys and profilesOfficial documentationdocs.mistral.ai
  9. Mistral docs: ConfigurationOfficial documentationdocs.mistral.ai
  10. Mistral docs: Using offline modelsOfficial documentationdocs.mistral.ai
  11. Mistral docs: Safety, approvals, and permissionsOfficial documentationdocs.mistral.ai
  12. Mistral docs: AgentsOfficial documentationdocs.mistral.ai
  13. Mistral docs: Admin configOfficial documentationdocs.mistral.ai
  14. Mistral docs: Use Vibe in other IDEsOfficial documentationdocs.mistral.ai
  15. Mistral docs: Vibe Code Web, get startedOfficial documentationdocs.mistral.ai
  16. Mistral docs: Regional inferenceOfficial documentationdocs.mistral.ai
  17. Mistral docs: Zero data retentionOfficial documentationdocs.mistral.ai
  18. Mistral help: Do you use my user data to train your models?Official documentationhelp.mistral.ai
  19. Mistral Vibe on GitHub (mistralai/mistral-vibe)Official documentationgithub.com
  20. Mistral Vibe releases on GitHubOfficial documentationgithub.com
  21. Mistral Vibe license (Apache 2.0)Official documentationgithub.com