Terminally online Mistral Vibe.
Mistral AI
Mistral just shipped Vibe 2.0, a terminal coding agent running on its new Devstral 2 models. It now asks before guessing, lets you build custom subagents, and Devstral 2 stops being free outside a limited plan.
Based on reporting by Mistral AI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Mistral is pushing further into the crowded terminal-coding-agent race with Vibe 2.0, and the pitch this time is control rather than raw horsepower. The tool already did the basics well: natural language commands, multi-file edits, full codebase context. What's new is the ability to shape how the agent behaves rather than just what it does.
The headline feature is custom subagents — you can spin up narrow, purpose-built agents for things like PR reviews or deploy scripts and call them whenever you need. Pair that with multi-choice clarifications, where Vibe stops and asks which option you meant instead of quietly assuming, and you get an agent that's less likely to bulldoze through ambiguous instructions. Slash-command skills let you preload workflows — linting, docs generation, deployment steps — so you're not retyping the same prompts every session. And unified agent modes bundle tools, permissions, and behavior into switchable configurations, which matters more than it sounds for teams juggling different codebases or client contexts.
The CLI itself now updates automatically, no more manual patching, which is a small but telling sign Mistral wants this treated like infrastructure, not a side experiment. Underneath it all sits Devstral 2, the coding model family Mistral positions as state of the art for this kind of agentic work.
Here's the catch: Devstral 2 is no longer free through the API. Pricing lands at $0.40 per million input tokens and $2.00 per million output tokens for the full model, with a cheaper Small variant at $0.10/$0.30. Free access survives only on the Experiment plan in Mistral Studio, aimed squarely at prototyping rather than production. Vibe itself ships on Le Chat Pro and Team plans, with pay-as-you-go overflow or bring-your-own-key options, and Mistral is layering enterprise add-ons — fine-tuning on internal DSLs, reinforcement learning on custom environments, full codebase modernization — clearly chasing large regulated clients in finance, defense, and infrastructure who need more than a chat window can offer.
My take — AI-written commentary, not fact-checked reporting
The customization stuff is genuinely useful, but let's be honest about what's happening: Mistral spent a year building goodwill with free Devstral access and is now converting that goodwill into a paid funnel, which is a completely normal and slightly cynical move every AI lab eventually makes. I like that a European lab is building serious dev tooling instead of just chasing chatbot benchmarks, and the enterprise angle — finance, defense, infrastructure — is where Mistral's sovereignty pitch actually has teeth against US-hosted alternatives.}}}}}}}}}}}}}}
Read more about this at: Mistral AI