Pydantic released pydantic-ai v2.48.0 on 2026-09-22, adding model support for OpenAI's gpt-6-sol and gpt-6-luna, Anthropic's claude-opus-5-5, and new OpenAI audio models via genai-prices 0.1.7. It also fixes message-ID preservation in Vercel AI and AG-UI adapters and bumps genai-prices to 0.1.8.
Pydantic released six consecutive pydantic-ai versions (v2.38.0–v2.43.0, 2026-09-03 to 2026-09-11). Highlights include support for OpenAI gpt-6-astra and gemini-3.8-flash, Claude Fable 5.1 and Mythos 5.1, a vLLM provider, a GitHub Copilot provider, ChatGPT/Codex subscription auth, context-window tracking, barge-in handling in realtime sessions, and an @agent.on_event listener API. The releases matter to practitioners building agents who need current model support, cost tracking, and realtime voice control.
Pydantic released six consecutive pydantic-ai versions (v2.38.0–v2.43.0, 2026-09-03 to 2026-09-11). Highlights include support for OpenAI gpt-6-astra and gemini-3.8-flash, Claude Fable 5.1 and Mythos 5.1, a vLLM provider, a GitHub Copilot provider, ChatGPT/Codex subscription auth, context-window tracking, barge-in handling in realtime sessions, and an @agent.on_event listener API. The releases matter to practitioners building agents who need current model support, cost tracking, and realtime voice control.
Pydantic released pydantic-ai v2.47.0 on 2026-09-21, adding TypeSafeModel routing improvements and stricter validation of user prompt content. The release includes compatibility notes that change route naming and error behavior.
Pydantic released pydantic-ai v2.47.0 on 2026-09-21, adding TypeSafeModel routing improvements and stricter validation of user prompt content. The release includes compatibility notes that change route naming and error behavior.
Pydantic released version 2.10.0 of pydantic-ai on July 14, 2026, adding support for OpenAI background mode and handling Anthropic's stop_reason=pause_turn. The release also fixes bugs in message_history validation and RenamedToolset name collisions.
Pydantic released version 2.10.0 of pydantic-ai on July 14, 2026, adding support for OpenAI background mode and handling Anthropic's stop_reason=pause_turn. The release also fixes bugs in message_history validation and RenamedToolset name collisions.
Pydantic released version 2.17.0 of pydantic-ai on 2026-07-23, adding support for arbitrary fields in RequestUsage and RunUsage to prepare for upcoming genai-prices features, and caching per-message OpenTelemetry serialization to reduce instrumentation cost from O(n²) to O(n). The release also fixes ContentFilterError handling for Model Armor response blocks.
Pydantic released version 2.17.0 of pydantic-ai on 2026-07-23, adding support for arbitrary fields in RequestUsage and RunUsage to prepare for upcoming genai-prices features, and caching per-message OpenTelemetry serialization to reduce instrumentation cost from O(n²) to O(n). The release also fixes ContentFilterError handling for Model Armor response blocks.
pydantic-ai released v2.35.0 on 2026-08-25, deprecating RunContext.capability_loaded and available_capability_ids in favor of capability_active and active_capability_ids, reducing default Temporal metric export frequency, and fixing tool-description handling. Three follow-up patch releases (v2.35.1 through v2.35.3) landed within days, fixing Bedrock guardrail traces, Temporal cancellation behavior, Heroku GLM model routing, and deferred-tool result resolution. Developers using capability APIs or durable execution backends should review the deprecations and pin updates.
pydantic-ai released v2.35.0 on 2026-08-25, deprecating RunContext.capability_loaded and available_capability_ids in favor of capability_active and active_capability_ids, reducing default Temporal metric export frequency, and fixing tool-description handling. Three follow-up patch releases (v2.35.1 through v2.35.3) landed within days, fixing Bedrock guardrail traces, Temporal cancellation behavior, Heroku GLM model routing, and deferred-tool result resolution. Developers using capability APIs or durable execution backends should review the deprecations and pin updates.