pydantic/pydantic-ai v1.74.0

v1.74.0 (2026-03-30)

on GitHub

What's Changed

🚀 Features

• feat: online evaluation infrastructure for pydantic-evals by @dmontagu in #4782
• feat: Add TextContent for user prompts with metadata not sent to model by @thisisarko in #4432
• Make run ID time-sortable; add agent name and run ID span attrs to all agent run child spans by @adriangb in #4876
• Add AbstractToolset.get_instructions method and include_instructions arg to MCP Servers by @DougTrajano in #4123
• deprecate: Dataset without name parameter by @Kludex in #4862

🐛 Bug Fixes

• Fix(tracing): prevent CallDeferred/ApprovalRequired from marking tool spans as errors by @jaiwinshah3 in #4661
• fix: service_tier validation failure for OpenAI-compatible providers by @majdalsado in #4911
• Fix batch embeddings generation with gemini-embedding-2-preview by @DouweM in #4913
• fix(output): respect ToolOutput.max_retries parameter by @JasonCZMeng in #4687

New Contributors

• @jaiwinshah3 made their first contribution in #4661
• @thisisarko made their first contribution in #4432
• @adriangb made their first contribution in #4876
• @JasonCZMeng made their first contribution in #4687

Full Changelog: v1.73.0...v1.74.0