Pydantic AI v2.35.0 — deprecates RunContext.capability_loaded/available_capability_ids; Temporal metric export throttled
Pydantic
v2.35.0 released Aug 25. Deprecates `RunContext.capability_loaded` and `available_capability_ids` in favor of new capability APIs. TestModel: inclusive integer `maximum` becomes reachable. Default Temporal metric export frequency reduced. Explicit empty `Tool` description preserved instead of falling back to function docstring. Follows hot on v2.34.0 (Aug 24) which added a LangChain migration skill, `ZaiModel` GLM-5.3 support, and ~13 bugfix items including `VercelAIAdapter` rejecting reasoning-part `id`, Bedrock `r1` alias recognition, and `VercelProvider` dropping the Groq profile.
Why it matters
The deprecation is the kind that breaks consumer code that introspects `RunContext` for capabilities. The follow-up to v2.34.0 (which had a LangChain migration skill and Z.ai GLM-5.3) shows the framework is rapidly consolidating two ecosystems (Pydantic-AI-native + LangChain-skills) — users tracking both should know the API surface is being re-shuffled.
Importance: 2/5
default