Mistral Medium (latest) (EU)

Mistral Medium (latest) (EU) sits in the mistral-medium family as a multimodal generalist, pairing a 131K-token context window with the ability to read text and images while returning text, which makes it a natural choice for document-heavy assistants, visual question answering, and structured extraction tasks. The model is offered as open weights, signaling that teams can self-host or fine-tune for specialized pipelines while still relying on hosted inference for production traffic. Practical integrations are well supported through tool calling, structured output, temperature control, and attachment handling, so developers can wire it into agent workflows, RAG stacks, and API-based automations without unusual plumbing.

Positioning within the broader mistral-medium line gives the EU variant a clear role as the latest flagship for routed serving, sitting alongside sister releases such as the 3, 3.1, 3.5, and 3.5-thinking editions and the non-EU Mistral Medium latest. Compared with those siblings, this EU-routed build keeps the same 131K context ceiling while distinguishing itself through regional hosting and routing through the Requesty endpoint, which can matter for data residency and latency inside Europe. For teams choosing among the medium-tier models, the latest EU version is best suited when multimodal input, open-weight flexibility, and standardized developer tooling matter more than the larger 262K contexts or extended thinking variants available elsewhere in the family.