Mistral has published version 2.10.0 of its official Python SDK, and the update matters more to teams maintaining administrative integrations than to developers who only call models. The package reorganizes the connector surface, adds credential and service account capabilities, and removes 19 methods tied to campaigns, chat events, datasets, and fields. This story is labeled CONFIRMED: the version was released in the official repository and is also available on PyPI.
The central change is in connectors. The create method now takes a single request object instead of accepting loose named parameters such as name, description, server, and title. Connector is also now a union of HTTPConnector and MCPConnector, discriminated by the protocol field. In TypeScript, protocol must be provided explicitly; in Python, the library still supplies a default. The design is less ambiguous, but it breaks code that builds connectors with the previous signature.
The release also makes access management more explicit. It adds operations to create, update, share, and unshare organization credentials, plus a group of endpoints for service accounts and assignable roles. For companies using Mistral in internal products, automations, or agents, the practical benefit is clearer separation between a human identity, a technical credential, and an organizational scope. For a small Brazilian company, that can mean less dependence on the founder’s personal key when a system becomes a team tool.
The tradeoff is immediate migration work. The release removes endpoints for campaigns, chat event search and evaluation, dataset imports, and field option queries. Mistral’s official migration documentation still treats chat, streaming, embeddings, agents, function calling, and batch as unchanged areas in the move to the version 2 line, but the release list shows that administration, evaluation, and observability capabilities are evolving. Teams that rely on them must review code and tests instead of merely upgrading the dependency.
There are also compatibility fixes that look small but prevent real failures. Transcription segment start and end fields are now nullable because the API can return missing values in diarized transcriptions. OAuth metadata, service tier information, and deployment worker responses changed as well. These are signs of an SDK tracking a broader platform, not a cosmetic patch.
For Brazilian small and midsize businesses, the advice is direct: pin the version in the environment, read the breaking changes, search for connectors.create calls, and test transcriptions with segments that have no timestamps. This release contains no evidence of a new model or a price cut. Its value is integration governance. My read is that Mistral is moving the SDK toward an architecture better suited to multi-user environments and MCP, while charging that evolution in client-side rework. Upgrading without a contract test is an invitation to find the break in production.
Sources
- Official Mistral Python client release: https://github.com/mistralai/client-python/releases/tag/v2.10.0
- mistralai 2.10.0 on PyPI: https://pypi.org/project/mistralai/2.10.0/
- Official Mistral SDK documentation: https://docs.mistral.ai/resources/sdks
- Official Python SDK v2 migration guide: https://docs.mistral.ai/resources/migration-guides