Mistral Vibe pricing: plan budgets, pay-as-you-go keys and your own provider

Mistral Vibe the CLI is free; what you pay for is the model, and Mistral bills it three different ways depending on which key Vibe holds. A plan — Free, Pro at $14.99 a month or Team at $24.99 per seat — gives a monthly Vibe budget of unpublished size. A pay-as-you-go key bills per token at Mistral's API prices, $1.5 in and $7.5 out per million for the default Mistral Medium 3.5. And a provider of your own bills you instead, except for a few background calls. Prices read from mistral.ai on October 1, 2026.

This page is about the money. For Vibe against Anthropic's agent see Mistral Vibe vs Claude Code; for other terminal agents see Mistral Vibe alternatives. Since May 28, 2026 "Vibe" is also the name of Mistral's whole assistant (formerly Le Chat), so a Vibe plan pays for chat, work and coding together — the terminal agent is one surface of it.

The plans, as they touch the CLI

Read from mistral.ai/pricing on October 1, 2026, prices before tax. Pro and Team are "subject to fair usage limits", and Mistral publishes no session count, token cap or dollar figure for any plan's Vibe budget, so no plan converts into hours or tokens. The last column is the comparison table's "Extend usage with PAYG credits" row: "Keep going past your plan limit by paying only for what you use beyond it at API rate."

Which key Vibe holds decides the bill

Vibe's setup page offers two access paths on first launch: sign in with a Mistral account and "use your Mistral plan", or paste an API key. Mistral's rate-limit article then separates the keys you can paste:

If you script Vibe, Mistral's own advice is the middle column: a quota email about a "Vibe Code CLI usage quota" usually means an automation was given a plan key. Prepaid credits are applied as your primary balance before your card is charged, but they do not raise the rate-limit tier — only billed usage does. Both plan and API usage show as a percentage under Code › CLI → View organization usage.

What a session costs at API prices

These are token arithmetic, not measured task costs and not a bill ceiling. One session that sends 200,000 uncached input tokens and receives 12,000 output tokens:

The decision the table supports: at Medium 3.5's list price, Pro's $14.99 equals about 38 sessions of that shape a month. Below that, a pay-as-you-go key is cheaper than Pro; above it, Pro's budget probably is, for as long as the budget lasts — and Mistral does not say how long that is. Caching moves the line a long way, and so does the model: Small 4 is a tenth of Medium's price and not the same agent. Cheapest listed rate and cheapest way to finish a task are different questions, because a model that needs three attempts can cost more than a dearer one that needs one.

The Medium and Small rates are from Mistral's API price list; the cached rate is the one Vibe's shipped configuration declares for its default model. Kunavo does not sell Mistral models; its rows are live catalog rates, and the catalog amount is a billing floor rather than a cap — when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup.

Two traps on your own provider

Background calls still go to Mistral. In Vibe's source, small utility completions — conversation titles, worktree names — go to Mistral's mistral-vibe-cli-fast model whenever a Mistral provider is usable, even when the session runs elsewhere, and fall back to your active model only when none is. So a session on your own provider can still draw a trickle from your Mistral plan or pay-as-you-go key while MISTRAL_API_KEY is set. Unset it, or exclude the model with allowed_models, to keep everything on one bill. Mistral Vibe vs Claude Code covers the same routing for Smart Approve and connectors.

--max-price sees only the prices you typed. In programmatic mode, vibe -p … --max-price 2 stops the session once its computed cost passes $2 — and that cost is computed from each model entry's input_price and output_price, which default to 0.0. A custom model declared without them costs "$0" forever and the cap never fires. Vibe's source also notes that switching model mid-session reprices every token at the new model's rate. Either declare the prices, or cap on --max-tokens, which needs none:

# ~/.vibe/config.toml — a custom model with prices, so --max-price can see it
[[models]]
name = "claude-sonnet-5"
provider = "kunavo"
alias = "sonnet-kunavo"
input_price = 1.4     # USD per million input tokens
output_price = 7    # USD per million output tokens

# Then, in programmatic mode only:
#   vibe -p "fix the failing test" --max-price 2
# Without the two price lines both default to 0.0, the session cost stays $0,
# and --max-price never stops anything. --max-tokens needs no prices.

The full provider block is on Kunavo's Vibe setup page. Kunavo bills per token from a prepaid balance with a $10 minimum top-up and no subscription; see billing. Nobody at Kunavo has run Vibe against its endpoint, so read the first request's charge in your usage log before a long session.

FAQ

Read on October 1, 2026: mistral.ai/pricing (both states of its student switch, and the plan comparison table), mistral.ai/pricing/api, docs.mistral.ai's Vibe Code overview and CLI setup pages, and Mistral's help-centre articles on rate limits, refunds and zero data retention; the mistral-vibe 2.25.8 wheel from PyPI (source of the price fields, the price cap and the utility-model routing). No Mistral account was used and Vibe was not run, so nothing here is a measured bill; every dollar figure is arithmetic on published rates.