Mistral AI review

Mistral prices per model rather than on one shared plan. Entry rates sit at input fees of $0.04 to $1.50 and output fees of $0.04 to $7.50 across the catalogue. Entry input price ranks 21 of 71 on the site, while output price ranks 10 of 70, so the output side is the stronger of the two.

Monthly API credits come with the Free plan, and the Education Pro plan carries more of them. That covers prototypes, side projects and coursework, while steady production traffic runs on metered rates of $0.04 to $1.50 for input and $0.04 to $7.50 for output. Credits are the entry route, not the whole pricing model.

The catalogue runs to Mistral Medium 3.5, Mistral Small 4 and Mistral Large 3, covering general chat models, small Ministral sizes, Codestral for code, OCR and transcription, embedding, moderation and Shieldstral, plus Z.ai GLM models carried alongside Mistral's own.

The largest context window is 1,000,000 tokens, which ranks 21 of 72 across the brands on the site and sits at the upper end of the field at 1,048,576. Long documents and big codebases fit without heavy chunking.

Support runs through help centre, community and email, covering self-serve articles, community answers, email and live chat. Phone lines are the exception across providers here, at 43.6% against 90.2% for help centres, so expect to resolve most questions in writing or chat.

Mistral is a simplified joint stock company (S.A.) registered in Paris, headquartered in France, and it was founded in 2023. Commercial terms and support come from that same entity, with help routed through help centre, community and email rather than a separate reseller.

A gateway publishes one input figure: Salad lists input price from $0.06, where Mistral's per-model input fees of $0.04 to $1.50 span a wider range. The site median entry input price is €0.07, so Mistral's cheapest tier sits under it while the top of its range sits over.

Supascale AI routes DeepSeek V4 Pro, GPT 5.6 Sol and GPT 5.6 Terra, and Mistral and Mixtral appear in that list, so a multi-lab gateway can serve the same models on different terms. That suits teams already buying across several labs and happy to route the same work through a shared endpoint.