Mistral Vibe

You will get a proper terminal agent, and Pro at $14.99 a month covers it in the CLI, the VS Code extension and the web surface at chat.mistral.ai/code, so the daily trait is one plan that follows you across three surfaces without a second login. Devstral is good, not the best, and on a hard refactor the gap shows in the number of retries.

For the routine half of the week that gap does not matter and the price does. Pick it for value and a European lab behind the model. Pick Claude Code when the task is the hard one and the bill is worth it.

Two lines from the configuration docs. First: programmatic mode falls back to auto-approve when --agent is not provided, so a script that forgets the flag runs every tool unasked, and a CI job is exactly where someone forgets a flag. Second: admin-managed config is a distribution mechanism, not a security control, which is honest and means a user can override it.

The consequence: treat every scripted invocation as unattended by default and pin the agent explicitly. A safe default for programmatic mode would change this verdict. What it does right: enabled_tools and disabled_tools accept globs and regex, so you can fence the agent per project.

Claims first. The 72.2% on SWE-bench Verified is reported by Mistral for the Devstral 2 model, with no harness disclosed, so it says nothing about the CLI and should not be quoted as an agent result. Architecture: 1. Read, patch and search tools. 2. A todo list as explicit plan state, visible to the user. 3. Subagents for delegation.

The consequence is that the number and the product are measured separately, and only the number has been. A run of the CLI on the same set would settle it. The observation: the benchmark measures the model, the product is the loop.

The rename of Le Chat to Vibe on May 28, 2026 says coding is now the flagship at Europe's frontier lab, and a lab that renames its consumer brand after a developer tool has told you where the revenue plan lives. Pricing power sits in the model: Devstral 2 at $0.40 and $2.00 per million tokens undercuts the American labs, a price war Mistral can afford only with sovereign customers behind it.

The likely path is sovereignty contracts, not acquisition, since the buyer list for a European lab is political. Position: safe for 18 months, the model is the product.

The demo is a terminal agent. Procurement: Team at $24.99 per user is $1,499.40 a month for sixty, Enterprise is custom, and nothing on the pricing page names SSO, SCIM or retention, which for a European vendor is a surprise, since that is the sales pitch. A headless mode makes CI reachable.

Onboarding is a curl install and a login, and the config file can be distributed centrally even if it cannot be enforced. Approved with conditions: Team now for the pilot, Enterprise once the SSO and retention terms are in writing, and the config distributed from one repo.

Apache-2.0 and the config is a TOML file I can version: ~/.vibe/config.toml, or ./.vibe/config.toml per project, with providers defined by api_base, api_key_env_var and api_style, so my local server is one block away and the vendor's model is another. MCP servers are [[mcp_servers]] entries with a transport and a url.

ACP mounts it in Zed or any editor that speaks it, so the CLI is the agent and the editor is a view. The pip install means I can patch site-packages when I disagree. A fork survives the vendor. Respect, no asterisk.