Mistral Pricing: Which Model Is Worth Paying For?
OpenRouter lists 19 Mistral models. For 1M input plus 1M output tokens they range from $0.049 (Mistral Nemo) to $9 (Mistral Medium 3.5), a 184x spread. Most teams overpay by defaulting to the flagship for work a cheaper model passes.
Plans and pricing
Mistral Nemo
$0.019 input / $0.03 output per 1M tokens131K context on OpenRouter (mistralai/mistral-nemo). 1M input plus 1M output tokens: $0.049.
Best for
High-volume, low-stakes steps: routing, tagging, extraction and first drafts.
Not for
Tasks where a wrong answer costs more than the tokens saved.
Mistral Small 4
$0.15 input / $0.60 output per 1M tokens262K context on OpenRouter (mistralai/mistral-small-2603). 1M input plus 1M output tokens: $0.75.
Best for
A middle tier to test before paying flagship prices.
Not for
Tasks where a wrong answer costs more than the tokens saved.
Ministral 3 8B 2512
$0.15 input / $0.15 output per 1M tokens262K context on OpenRouter (mistralai/ministral-8b-2512). 1M input plus 1M output tokens: $0.30.
Best for
A middle tier to test before paying flagship prices.
Not for
Tasks where a wrong answer costs more than the tokens saved.
Mistral Medium 3.5
$1.50 input / $7.50 output per 1M tokens262K context on OpenRouter (mistralai/mistral-medium-3-5). 1M input plus 1M output tokens: $9.
Best for
The hardest Mistral tasks, once cheaper Mistral models have failed your acceptance tests.
Not for
Bulk work a smaller model already passes; the price gap compounds at volume.
Prices are OpenRouter's listed per-token prices for Mistral AI models, read from the OpenRouter models API and checked 2026-10-04. Buying direct from Mistral AI or another host can be priced differently; prices exclude tax and change often.
Who should pay for Mistral?
Mistral suits European buyers and teams that want small, fast models for routing, classification and drafting. Start with Mistral Nemo on a sample of real tasks, measure the acceptance rate, and step up a tier only where it fails. Price the whole job, not the token: retries and human review often cost more than the model.
Calculate your exact cost →Who is overpaying for Mistral?
- Teams sending every request to Mistral Medium 3.5 when Mistral Nemo passes most of them
- Anyone paying for output tokens they never read: cap max tokens and ask for concise answers
- Buyers comparing token prices without counting retries, failed responses and review time
Official pricing sources
Frequently asked questions
What is the cheapest Mistral model?
On OpenRouter, Mistral Nemo at $0.019 input and $0.03 output per 1M tokens (checked 2026-10-04).
How much does the top Mistral model cost?
Mistral Medium 3.5 is listed at $1.50 input and $7.50 output per 1M tokens, so 1M of each costs $9.
Is Mistral cheaper than GPT or Claude?
Usually per token, yes. Compare on your own prompts with the AI Cost Calculator: the cheaper model only saves money if it passes the same acceptance tests.