benchrPricing breakdown
| Tier | Rate / 1M tokens |
|---|---|
| Standard input | $1.50 |
| Standard output | $7.50 |
| Context window | 256,000 tokens |
Vision and PDF: what it enables
Mistral's model card lists multimodal use, Document Q&A, OCR, structured annotations, and bounding-box extraction for Medium 3.5. Those capabilities are exposed through documented chat, conversations, and OCR endpoints rather than one universal payload. Potential workflows include invoice extraction, document review, form processing, and image-plus-text analysis, but confirm the endpoint, supported formats, and extraction quality for the exact task.
Mistral Medium vs Mistral Large: compare both sides of the rate card
Mistral Medium 3.5 is listed at $1.50/1M input and $7.50/1M output. Mistral Large 3 is listed at $0.50/1M input and $1.50/1M output, so Medium's published rates are 3× higher for input and 5× higher for output. Price does not establish task quality: test both models on the same documents, tool calls, and output criteria before selecting one.
Hosted-data location is configurable, not automatic compliance
Mistral says hosted data is in the European Union by default and that an explicitly selected US API endpoint stores data in the United States. Its help center also says some features can temporarily transfer data outside the EU through listed subprocessors, with safeguards such as Standard Contractual Clauses where applicable. That makes region and feature configuration relevant, but it does not by itself establish GDPR or sector-specific compliance; review the current DPA, subprocessors, retention settings, and your legal requirements.
Modified MIT license: check the revenue condition
Mistral publishes Medium 3.5 as open weights under its Modified MIT license. Mistral's license guidance says companies with monthly revenue above $20 million must either obtain a commercial license or use the model through Mistral Studio. This summary is not legal advice; read the model's current license text before self-hosting or redistribution.
Cost scenarios
At 5M input + 1M output per month, the published rates produce $7.50 + $7.50 = $15/month. At 10M input + 5M output, they produce $15 + $37.50 = $52.50/month. These are rate-card scenarios only and exclude tools, storage, hosting, taxes, and any negotiated terms.
Use-case fit
Candidate for: Multimodal document and image workflows that need a 256K context window; teams already using Mistral's API; or self-hosting evaluations where Modified MIT's terms fit the organization.
Compare alternatives if: Your workload is price-sensitive, your organization exceeds the license's revenue threshold, or your governance requirements depend on a specific region, retention policy, subprocessor list, or certification.
Decision checklist
Confirm the API region and every enabled feature rather than treating the provider's headquarters as a compliance result. Review Mistral's current DPA, subprocessor disclosures, and retention controls with the people responsible for privacy and security.
For self-hosting, read the Modified MIT text and apply its revenue condition to the correct company entity. For API use, price a representative input/output mix and validate task quality on held-out examples.