Mistral Cost Calculator: estimate Mistral AI API spend
Estimate Mistral AI API spend across Mistral Large 2, Medium 3, Small 3, and Codestral. Enter input/output tokens (or paste a prompt) and your monthly call volume — get per-call, per-day, per-month and per-year cost in USD, EUR or GBP.
Mistral API pricing reference (2026)
Current per-million-token rates published on the provider's website. Toggle "Edit rates" above if you have a negotiated rate or want to model future pricing.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| Mistral Large 2 Mistral AI | $2.00 | $6.00 | 128K |
| Mistral Medium 3 Mistral AI | $0.40 | $2.00 | 128K |
| Mistral Small 3 Mistral AI | $0.10 | $0.30 | 32K |
| Codestral Mistral AI | $0.20 | $0.60 | 256K |
What you'll use this for
LLM costs scale fast. A quick estimate before you ship saves real money in production.
Budget planning
Project monthly and annual Mistral API spend before you ship. Catch the surprise before billing does.
Tier comparison
Small 3 vs Medium 3 vs Large 2 — see when each tier earns its keep on cost-per-quality.
Codestral planning
Estimate spend for repo-wide refactors and PR reviews using Codestral's 256K context.
Cost optimization
Toggle Advanced rates to model batch discounts, caching savings, or negotiated enterprise pricing.
How to estimate Mistral cost
Pick a model & enter tokens
Either paste a real prompt (auto-counted) or type input tokens directly. Set expected output tokens for the model's reply.
Set monthly call volume
Drag the slider. The tool projects per-call, per-day, per-month and per-year cost in your selected currency.
Compare or override
Switch the model dropdown to compare tiers. Toggle Advanced to override rates if you have a custom contract.
Frequently asked questions
From Mistral AI's la Plateforme pricing page as of 2026. The figures shown are list rates for the standard chat completion endpoint. Self-hosting on your own GPUs or via Bedrock/Azure has different cost dynamics — verify with the relevant provider before procurement.
At $0.40 input and $2.00 output per million tokens, Medium 3 is roughly 5× cheaper than Mistral Large 2 while delivering 90%+ of the quality on most evals. It's the new default recommendation for teams who don't strictly need frontier reasoning.
Mistral offers a batch API at roughly 50% off list prices for offline workloads. Prompt caching is available on supported tiers with cache-read pricing about 10× cheaper than fresh-input pricing. Toggle Advanced above to model these scenarios with your own override rates.
Codestral is a smaller, specialized model — 22B parameters dedicated entirely to code. Mistral can offer it at $0.20/$0.60 because the inference cost is genuinely lower than the general-purpose Medium 3. The 256K context comes from architectural choices (sliding-window attention) that don't scale serving cost linearly.
About Mistral pricing
Mistral AI's 2026 pricing is the most aggressive of any frontier-class provider — Medium 3 in particular is positioned as the price/quality leader for general-purpose work. The whole family bills per million tokens with no minimum spend.
Per-tier breakdown
- Mistral Large 2 — $2.00 input / $6.00 output per M. Flagship multilingual reasoning, competitive with Sonnet 4.6 at lower price.
- Mistral Medium 3 — $0.40 input / $2.00 output per M. The new default. 5× cheaper than Large 2 with 90%+ quality on most benchmarks.
- Mistral Small 3 — $0.10 input / $0.30 output per M. Latency-optimised 24B model — great for chat, classification, RAG.
- Codestral — $0.20 input / $0.60 output per M. Code specialist with 256K context, fill-in-the-middle, 80+ languages.
How costs scale
Output tokens dominate the bill on chat workloads. For Mistral Medium 3 at standard rates, a 500-token reply to a 200-token prompt costs about $0.0011 — multiply by call volume to see monthly impact. At 100K calls/month, that's ~$110.
When Mistral makes sense
Multilingual work (Mistral excels at European languages), code (Codestral is competitive with the best code models), price-sensitive volume (Small 3 and Medium 3 undercut almost everyone), and EU data-residency requirements (Mistral is a Paris-based EU provider, an important compliance factor for European customers).