Mistral Medium 3

Mistral Medium 3 (mistral) costs $0.40 per 1M input tokens and $2.00 per 1M output tokens, as last published by mistral.

Mistral Medium 3 has a context window of 131,072 tokens (the maximum combined input and output it can handle in a single request).

Mistral Medium 3 sits in the medium size tier. Size tiers run from micro (small enough to run on a device) through small, medium and large to extra large (frontier flagships), and are assigned from the model's family, naming and limits rather than its price.

Mistral Medium 3 can generate up to 131,072 tokens in a single response. Reasoning (thinking) tokens count toward this limit.

Mistral Medium 3's training data runs through May 2025, so it may not know about later events. Coolhand records the cutoff at month precision, since providers often publish only the month.

Mistral Medium 3 accepts image and text as input and produces text as output, as recorded in the Coolhand catalog.

No — the weights of Mistral Medium 3 have not been released, so it is only available through hosted APIs.

Yes — mistral deprecated Mistral Medium 3 on August 31, 2026. Deprecated; retires 2026-08-31. Replacement: Mistral Medium 3.5.

mistral retired Mistral Medium 3 on August 31, 2026.

Since Mistral Medium 3 is deprecated, mistral customers are moving to Codestral 25.01, Ministral 3B, and Mistral Medium 3.

Call GET https://coolhandlabs.com/api/v2/inference_apis?q[source_api_eq]=mistral&q[model_eq]=mistral-medium-2505 — free, no API key required, and scoped to just this model since source_api and model together are unique. See https://coolhandlabs.com/docs for the full REST API reference.

Coolhand also tracks Devstral 2, Devstral 2 (latest), Open Mistral Nemo, Codestral 25.01, Ministral 3B, and Mistral Medium 3 from mistral. Some may already be deprecated, so check each model's status alongside its current pricing.