Our take
Mistral AI is a France-based provider that exclusively hosts its own family of models, combining strong compliance credentials with measured throughput data. It offers an EU-region endpoint and an independent security attestation alongside a 12-offer sample with speeds ranging from 13 to 102 tokens per second.
Use this for EU data sovereignty, with France headquarters and a verified EU endpoint. Pick it when throughput matters — the fastest model hits 102 tokens per second. Choose it for first-party model fidelity. Skip it if you need non-Mistral models, if output-price premiums on larger models hurt your budget, or if verified zero-retention is mandatory.
- Strongest compliance posture in the sample: France headquarters, EU endpoint, no training on prompts, and SOC 2 attestation.
- Fastest measured throughput on a mid-size model — Mistral Small 3.2 24B at 102 tokens per second, 2.4× the next fastest in its size band and 7.8× the slowest in sample.
- Predictable one-to-one input and output pricing on entry models: Ministral 3.3B, Ministral 3.8B, and Mistral Nemo.
- Premium output pricing escalates sharply on larger models: output cost on Devstral 2 and Mistral Medium 3 is twenty times the entry level.
- Slowest throughput on code-specialized and largest models — Codestral 2508 at 13 tokens per second and Saba at 17.5 tokens per second, both below twenty.
- Narrow catalogue with no third-party models: all 17 tracked offers are from the Mistral family, with no Gemma, Llama, DeepSeek, or other alternatives.
Privacy & data handling
- ✓
- Yes
- ✗
- No
- No answer on record
The colour says whether the answer favours you, not whether it is a yes — not training on your prompts earns a green cross, no zero-retention option earns a red one.
Prompt retention: 30 days.