Summary: Mistral released Mistral Vibe 2.0 on January 27, 2026, moving its terminal coding agent from an open experimental CLI into paid Le Chat Pro and Team plans. Version 2.0 adds custom subagents, clarification choices, slash-command skills, unified agent modes, and continues running on the Devstral 2 family. Fine-tuning, self-hosting, and BYOK aim the product at enterprises that prioritize control and data boundaries.
Mistral put out “Terminally online Mistral Vibe 2.0” on January 27, 2026 — a sweeping revision of its terminal-native coding agent, which now runs on the company’s Devstral 2 model family. Nothing about the release is ambiguous: Vibe has left behind the open-source experimental CLI first shown in December 2025 and now sells as a paid product packaged into Le Chat Pro and Team plans. VentureBeat read the launch as a European challenger going after GitHub Copilot.
When it arrived matters too. Claude Code, OpenAI Codex, and a short list of others hold the terminal coding agent category today, and Mistral stands among the few European model vendors actually in the fight. Qwen shipped Qwen3-Max-Thinking just one day earlier, and this week’s pile-up of product launches makes the point: coding agents are where 2026 opens.
Out of the Experiment and onto the Price List
Vibe 1.0 arrived on December 9, 2025, released together with Devstral 2 and described as an open-source command-line assistant that explores a codebase, modifies it, and executes the changes. Version 2.0 carries forward the original toolkit — natural-language commands, multi-file orchestration, smart references, full codebase context — and bolts on auto-updates, so CLI fixes arrive continuously instead of waiting on a manual upgrade.
The deeper break with version one is commercial. Per The Decoder, Vibe 2.0 ships inside Le Chat Pro at $14.99 a month and Team plans at $24.99 for each seat; once usage outgrows the plan, you either meter at API rates or bring your own API key (BYOK). Students pay 50% less for Pro. Setup stays minimal: pull the CLI from the mistral-vibe package via uv or the install script, register at Mistral’s console, and run it.
The Four Additions in Version 2.0
- Custom subagents: create purpose-built agents for specific jobs — deploy scripts, PR reviews, test generation — and call them up whenever needed.
- Multi-choice clarifications: in Mistral’s words, “Vibe asks before it acts.” Given an ambiguous instruction, it lays out the options rather than taking a guess.
- Slash-command skills: a ”/” call brings up preconfigured workflows for jobs like deploying, linting, or generating documentation.
- Unified agent modes: tools, permissions, and behaviors get packaged into modes, so switching context never requires switching tools.
Beneath all four sits one design judgment: once engineering work stretches long and spans many steps, keeping control matters more than “doing it all in one shot.” That stance separates 2.0 from agents chasing autonomy at any price, and lines up with the way conservative enterprise engineering teams actually prefer to hand off work.
Devstral 2 Goes Paid: Reading the Pricing Logic
Devstral 2 changed lanes at the same moment: open weights anyone could call became a paid API costing $0.40 per million input tokens and $2.00 per million output tokens, while the smaller Devstral 2 Small lands at $0.10 and $0.30. The one free path left is the Experiment plan inside Mistral Studio. Coverage from The Decoder distills Devstral 2’s edge into a cost position roughly seven times better than Claude Sonnet’s — and in agentic workflows that bill by the token and often stretch across dozens of steps, that edge lands directly on the invoice.
The Enterprise Strategy of a European Challenger
The enterprise side got equal airtime: Mistral described fine-tuning on customers’ internal languages and domain-specific languages, reinforcement learning run against each customer’s own environment, and “code modernization” — entire codebases relocated onto modern stacks with their business logic preserved. Large organizations in finance, defense, and infrastructure have already received these services, the company says. For European enterprises working under data sovereignty and compliance rules, a French vendor whose coding agent can be fine-tuned and self-hosted is an option that gets through internal policy review far more easily than the rest.
The Developer Takeaways
Three things worth carrying away. One: terminal coding agents now compete less on raw model capability and more on workflow integration — subagents, skills, and modes all exist to seat the agent inside the process a team already runs. Two: the price war has reached coding models, and Devstral 2’s rates make the tiered playbook — a cheap model carrying the many agent steps, a frontier model reserved for the decisive ones — genuinely practical. Three: BYOK and self-hosting show where Mistral has placed its wager: enterprises want control before anything else. Enterprise adoption numbers over the next few quarters will tell whether that wager pays.