Pydantic AI v2.45.0 (2026-09-17)
Pydantic AI v2.45.0 adds TypeSafeModel for TypeSafe's Jev, expands Bedrock Converse support for new models and high effort settings, and delivers multiple fixes for durable runs, MCP handling, and span usage reporting.
Why this signal matters
Pydantic AI version 2.45.0 introduces the TypeSafeModel supporting TypeSafe's Jev. Key bug fixes and optimizations include: passing 'xhigh' effort through Bedrock when supported by the model profile; enabling gpt-5.6-sol, gpt-5.6-luna, and gpt-5.6-terra on Bedrock Converse; performance/lifecycle enhancements in durable execution by resolving DynamicToolset and maintaining MCP server sessions once per durable run rather than per durable unit; preserving tool history inside MCPSamplingModel; fixing import issues with legacy httpx under FastMCP 4 setups in pydantic ai.mcp ; and isolating token/usage reporting to each agent run's respective span.
Actionable summary
Pydantic AI v2.45.0 introduces TypeSafeModel for Jev, fixes Bedrock Converse integration (supports gpt-5.6-sol/luna/terra and passing 'xhigh' effort), optimizes durable runs to resolve DynamicToolset and hold MCP server sessions once per run, fixes MCPSamplingModel tool history preservation, removes legacy httpx imports in MCP, and updates span usage reporting per agent run.
- Agent usefulness
- 82/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade to pydantic-ai v2.45.0 to resolve MCP session reuse and DynamicToolset resolution overhead in durable runs.
- Verify compatibility if using FastMCP 4, as legacy httpx imports have been removed from pydantic ai.mcp.
- Update Bedrock Converse configurations if adopting gpt-5.6-sol, gpt-5.6-luna, gpt-5.6-terra, or 'xhigh' effort parameters.