Mistral Vibe 2.0 adds subagents, slash-command skills and clarification prompts
Mistral AI
- Agents
- Models
- Developer tools
- #devstral 2
- #le chat
- #subagents
Mistral announced Vibe 2.0, an upgrade to its terminal coding agent built on the Devstral 2 model family. According to the announcement, it adds custom subagents, multi-choice clarifications before execution, slash-command skills, unified agent modes and automatic CLI updates. Mistral says Vibe is available on Le Chat Pro and Team plans with pay-as-you-go credits or bring-your-own-key, and that Devstral 2 API access has moved to paid pricing with a free Experiment plan.
What Mistral announced
Mistral published a release note for Vibe 2.0, described as an upgrade to its terminal-native coding agent and powered by the Devstral 2 model family. The stated additions are custom subagents for targeted tasks such as deploy scripts, PR reviews and test generation; multi-choice clarification prompts when intent is ambiguous; skills loaded with slash commands for workflows like deploying, linting or generating docs; unified agent modes combining tools, permissions and behaviors; and continuous CLI bug-fix updates without manual installs. The post also says Devstral 2 API access has moved to paid pricing, listing per-million-token rates and retaining free usage on an Experiment plan. The supplied excerpt supports what was announced and how Mistral prices it; it does not provide measured performance or independent evaluation of the new controls.
Scenario: a platform engineer standardizing PR checks
A platform engineer who already uses a terminal coding agent could trial a read-only subagent scoped to reviewing pull requests, keeping write actions out of its permissions at first. The premise that subagents can be built for tasks like PR reviews comes from Mistral's announcement; whether a given team's review conventions can be encoded well enough to be useful is an open question that the trial should answer. Start with one repository and a narrow review checklist rather than the whole monorepo.
Trial steps and success measures
A team could run a two-week comparison on one repository: route a fixed set of pull requests through the subagent and compare findings against the team's existing review baseline, recording missed issues, false positives and reviewer time. Add slash-command skills for one repetitive task, such as lint fixes, and log how often the clarification prompt interrupts versus prevents a wrong action. Failure checks should include stopping the trial if the subagent proposes changes outside its scope or if clarification prompts are ignored. Keep a human reviewer as the decision-maker throughout, and review logs before widening permissions.
Trade-offs, prerequisites and unknowns
Custom modes and subagents require upfront configuration and ongoing ownership so they do not drift from team conventions. The paid Devstral 2 API rates make cost monitoring a prerequisite; a team should estimate token use on its own workload rather than assume savings. Clarification prompts may add friction to fast workflows. The excerpt offers no evidence on output quality, latency or how well the clarification feature reduces errors, so it should not be treated as a guarantee. Organizations needing audited, vendor-managed tooling or with strict data-handling rules should not adopt on the basis of this announcement alone.
Context and limitations
- Based on a short excerpt of the Mistral announcement; no hands-on testing or independent benchmarks were reviewed.
- Pricing and plan details are as stated by Mistral and may change; verify current terms before budgeting.
- The excerpt does not quantify quality, speed or cost savings from the new features.
- Based on bounded source text, not independent testing. Practical suggestions are editorial proposals.