Quick answer
Mistral AI pricing on La Plateforme follows a pay-per-token model with separate input, cached-input and output rates. At standard regional inference, Mistral Small 4 costs $0.15 per million input tokens and $0.60 per million output tokens, while Mistral Large 4 costs $1.36 per million input tokens and $4.18 per million output tokens. The table above lists selected Mistral models. The Batch API offers lower rates for asynchronous jobs; check the official pricing page for model-specific batch rates.
The pricing shape of La Plateforme
La Plateforme prices per million tokens, quoted separately for input, cached input and output. Output is usually more expensive because it is generated one token at a time, but the gap varies by model — some Ministral models charge the same for input and output, while Medium 3.5 charges five times more for output. There is no fixed ratio; read each model's line on the official price sheet.
Where the money actually goes
- Input tokens: system prompt + user turns + tool JSON + retrieved context. Long RAG contexts dominate here.
- Cached input tokens: repeated prompt prefixes billed at the lower cached-input rate where Mistral applies it.
- Output tokens: the model's reply. Terse system prompts ("reply in one sentence") save real money.
- Embeddings: billed per million input tokens; there is no output side.
- Batch API: asynchronous requests can be cheaper than live inference; confirm the rate for your model on the official pricing page.
- Fine-tuning: billed separately; see the official pricing page for training and tuned-model rates.
Estimating a Mistral workload before you commit
Work from measured usage rather than estimates. Pull your input, cached-input and output token counts from the La Plateforme dashboard or your own request logs for a representative period, then apply the per-million rates in the table above to each figure separately. If you don't have measured usage yet, send a few real requests first and read the usage field from each API response — that gives you actual input and output token counts to extrapolate from. Output can be the bigger surprise: model replies and any tool_call JSON are billed in full, so measure it directly instead of assuming a ratio. A 30% buffer is an illustrative assumption for a first budget, not a universal recommendation — size your own buffer from your retry rate and traffic variability.
Levers that reduce your Mistral bill
- Structure prompts so long, reusable content comes first; where cached-input pricing applies, repeated prefixes cost a fraction of the standard input rate.
- Use the Batch API for anything that does not need to be live (evals, back-fills, offline scoring).
- Route easy turns to Mistral Small 4; escalate to Mistral Large 4 only when a cheap classifier says the turn is hard.
- Truncate RAG contexts aggressively — most passages contribute nothing above top-k = 5.
- Cap max_tokens on outputs; a runaway reply is the single most common surprise on the monthly invoice.
- Instrument per-endpoint token usage in your logs so you can spot the outlier route that eats 40% of the bill.
Batch API economics
The Batch API accepts JSONL files of requests and processes them asynchronously at lower cost than live inference. If a workflow tolerates async — nightly summarization, back-fills, evaluation runs — consider batching it. Batch availability and rates are model-specific, so confirm them on the official pricing page before you budget around them. Batch, cached-input and any negotiated rates should not be assumed to stack automatically.
Embeddings: Mistral Embed and Codestral Embed
Embeddings are billed per million input tokens with no output side: Mistral Embed costs $0.10 and Codestral Embed costs $0.15 per million input tokens. If you are recomputing an index on every deploy, cache aggressively — a document that has not changed does not need a new vector. Check the official documentation for which embedding endpoints support batch processing.
Comparing against other providers
Prices across LLM providers change frequently, so this guide does not quote competitor rates. Re-price your own workload on real traffic against each provider's official pricing page every couple of months. Mistral's EU-hosted default is a genuine differentiator for teams with data-residency constraints.
Reading your invoice
The console breaks down usage by API key, by model, and by request type. Tag your keys per environment (dev/stage/prod) and per feature so surprise bills are attributable in one click. Set spending caps in the console the day you take a payment method live — this is the cheapest insurance you'll ever buy.
EU-hosted defaults and regional pricing
The prices on this page are for standard regional inference. Mistral's default hosting is in the EU, which for many customers is a compliance win rather than a pricing lever. Egress costs, cross-region latency, and legal-review time all count against a workload's real total cost. If your app already runs on EU infrastructure, you get low-friction data residency; if it runs in US-east, add some round-trip time to your latency model.
Instrumenting token usage in production
Every Mistral response includes a `usage` block with prompt_tokens, completion_tokens, and total_tokens. Log all three per request, tagged with the endpoint that generated it, and pipe them into whatever observability stack you already use. Within a week you'll know which product surface is eating your budget and whether Small-tier routes are quietly escalating to Large in production without your knowing. This is the single most useful piece of infra you can add before your first invoice arrives.
When to talk to Mistral sales
If your usage is large and predictable, it can be worth contacting Mistral sales to ask about enterprise terms such as higher rate limits or support. Bring a clean usage report — the same one your invoice-attribution work produces. Any negotiated terms are set by Mistral case by case; this guide does not quote thresholds or discount levels.
Worked cost example: 1M input, 1M output tokens
Using the list prices in the table (uncached input, standard regional inference): on Mistral Small 4, one million input tokens costs $0.15 and one million output tokens costs $0.60, for a total of $0.75. The same volume on Mistral Large 4 costs $1.36 + $4.18 = $5.54. Scaling up, a workload doing 500M uncached input and 200M output tokens per day costs $75 + $120 = $195 per day on Mistral Small 4, and $680 + $836 = $1,516 per day on Mistral Large 4. Cached input and batch processing can lower these figures where they apply, but the math above assumes neither.
Batch vs realtime pricing decisions
Any workload that tolerates asynchronous turnaround is a candidate for the Batch API. That includes nightly report generation, evaluation runs across a golden set, back-fills after a schema change, and offline scoring of historical data. Realtime pricing only makes sense for turns that a user is actively waiting on. A common cost mistake is running evaluation jobs against the live endpoint because it is easier to script. Structure your codebase so switching a call site from live to batch is a one-line change; you will use it more often than you expect.
Tier recommendation by company stage
Early stage: stay on Mistral Small 4 for user-facing turns, use the Batch API for async work, and hold off on fine-tuning — a good system prompt is cheaper. Growing: introduce the Small-first, escalate-to-Large pattern, add per-endpoint token instrumentation, structure prompts for cached input, and request quota bumps proactively. Large, steady usage: talk to Mistral sales, consider fine-tuning for your top use cases, run a monthly cost review with per-team attribution, and evaluate self-hosting open-weight releases where data residency or economics justify the ops overhead.
How to estimate your monthly spend
Pull one week of production logs. For each endpoint that hits Mistral, compute mean input tokens, mean output tokens, and requests per day. For each row, cost per day equals (mean_input × input_rate + mean_output × output_rate) × daily_requests ÷ 1,000,000. Sum across endpoints, multiply by 30, and add a 30% safety margin for growth and variance. Split the total by model so you can see where the escalation architecture is (or is not) paying off. Re-run monthly — pricing changes, and workload mix shifts more than teams expect.