On January 27, 2026, Mistral shipped “Terminally online Mistral Vibe 2.0,” a major upgrade to its terminal-native coding agent, now powered by the company’s Devstral 2 model family. The signal in this release is unambiguous: Vibe has gone from an open-source experimental CLI, first shown in December 2025, to a paid product bundled into Le Chat Pro and Team plans. VentureBeat framed the launch as a European challenger going after GitHub Copilot.
The timing matters. Terminal coding agents are a category currently led by Claude Code, OpenAI Codex, and a handful of others; Mistral is one of the few European model vendors actually competing here. And with Qwen shipping Qwen3-Max-Thinking the day before, the density of product launches in this single week says coding agents are the opening battlefield of 2026.
From Experimental CLI to Paid Product
The first version of Vibe appeared on December 9, 2025 alongside Devstral 2, positioned as an open-source command-line assistant that explores, modifies, and executes changes across a codebase. Version 2.0 keeps the carried-over features — natural-language commands, multi-file orchestration, smart references, full codebase context — and adds auto-updates, with CLI fixes shipped continuously rather than through manual upgrades.
The real change is commercial. According to The Decoder, Vibe 2.0 comes with Le Chat Pro at $14.99 per month and Team plans at $24.99 per seat; beyond plan limits you either pay as you go at API rates or bring your own API key (BYOK). Students get 50% off Pro. Getting started stays lightweight: the CLI installs from the mistral-vibe package via uv or the install script, and after signing up at Mistral’s console you just run it.
Four New Capabilities
- Custom subagents: build specialized agents for targeted work — deploy scripts, PR reviews, test generation — and invoke them on demand.
- Multi-choice clarifications: Mistral’s phrasing is “Vibe asks before it acts.” When intent is ambiguous, it presents options instead of guessing.
- Slash-command skills: load preconfigured workflows with “/” for tasks like deploying, linting, or generating documentation.
- Unified agent modes: bundle tools, permissions, and behaviors into modes, so you switch contexts without switching tools.
All four point to the same design judgment: for long-horizon, multi-step engineering work, controllability beats “do it all in one shot.” It is a notably different emphasis from agents that optimize for autonomy at all costs, and it maps cleanly onto how conservative enterprise engineering teams actually want to delegate work.
The Pricing Play Behind Devstral 2’s Paid API
Alongside Vibe 2.0, Devstral 2 moved from freely callable open weights to a paid API: $0.40 per million input tokens and $2.00 per million output tokens, with the smaller Devstral 2 Small at $0.10 and $0.30. Free usage now survives only on the Experiment plan inside Mistral Studio. The Decoder’s related coverage summarizes Devstral 2’s edge as roughly a sevenfold cost advantage over Claude Sonnet — and for agentic workflows that bill by the token and routinely run dozens of steps, that is a number that shows up directly on the invoice.
A European Challenger’s Enterprise Play
Mistral also detailed its enterprise services: fine-tuning on internal languages and domain-specific languages, reinforcement learning against a customer’s own environment, and “code modernization” — migrating entire codebases to modern stacks while preserving business logic. The company says it has delivered these for large organizations in finance, defense, and infrastructure. For European enterprises bound by data sovereignty and compliance rules, a French vendor offering a coding agent whose model can be fine-tuned and self-hosted is an option that clears internal policy far more easily than most alternatives.
What It Means for Developers
Three observations. First, the competitive axis for terminal coding agents is shifting from raw model capability to workflow integration — subagents, skills, and modes are all techniques for embedding an agent into a team’s existing process. Second, the price war has reached coding models: Devstral 2’s rates make a tiered strategy — cheap model for the many agent steps, frontier model for the critical ones — genuinely practical. Third, BYOK and self-hosting signal that Mistral is betting enterprises want control above all. Whether that bet pays off will show up in enterprise adoption numbers over the next few quarters.
Sources
- Terminally online Mistral Vibe 2.0 — Mistral AI
- Mistral AI launches terminal-based coding agent Vibe 2.0 — The Decoder
- A European AI challenger goes after GitHub Copilot: Mistral launches Vibe 2.0 — VentureBeat
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.