pydantic-ai Releases

Introduces a new `XaiModel` using the xAI SDK while deprecating the older `GrokProvider` that used the OpenAI-compatible API. Also bumps dependencies.

Jan 22, 2026

Adds support for VoyageAI embeddings and fixes critical bugs including OpenRouter behavior changes for `DocumentUrl` and Gemini timeout issues.

Jan 22, 2026

Adds Exa search tools integration and support for Bedrock Nova 2.0's built-in Code Interpreter tool. Also fixes `count_tokens()` for `WrapperModel`.

Jan 17, 2026

Adds support for Google embedding models and fixes a bug where OpenAI was not populating tool arguments correctly.

Jan 15, 2026

Adds SambaNova provider support, consistently raises `ContentFilterError` on empty responses, and adds OpenTelemetry GenAI semantic conventions. Also fixes tool call cancellation.

Jan 14, 2026

Adds YAML and TOML media type support in `BinaryContent.from_path` and introduces metadata support for `DeferredToolResults.

Jan 10, 2026

Improves Temporal activity summaries with human-readable text and allows agents to call nonexistent tools up to the specified retries limit.

Jan 7, 2026

A patch release fixing `partial_output` flags in `TextOutput` during streaming, fixing duplicated tool calls, updating `thoughtSignature` for Vertex Gemini 3, and correcting `input_tokens` calculation for Bedrock models.

Jan 6, 2026

This release introduces support for embedding models and exposes agent and run metadata on result objects and span attributes. It also enhances system prompt handling and adds support for `ThinkingPart` in MCP Sampling.

Dec 24, 2025

Updates include adding local timestamps to request and response models and fixing media type inference for URLs with query parameters. It also improves error handling for AG-UI and adds documentation for Vertex AI.

Dec 23, 2025

This release enhances Temporal Agent capabilities by allowing model switching at runtime and supports new image generation options for Vertex AI. It also fixes a duplicate message issue in streaming output.

Dec 20, 2025

The focus is on AI SDK integration, specifically adding data chunk IDs and tool approval types. It also updates the Google Genai library and fixes schema transformer issues.

Dec 19, 2025

Introduces a `FileSearchTool` and support for new AI models including Alibaba Cloud DashScope and Gemini 3 Flash. It also adds AG-UI multi-modal message support and OpenAI reasoning options.

Dec 17, 2025

Adds a Web Chat UI for agents and improves file handling for Anthropic and OpenAI models. It also fixes compatibility issues with Temporal Agent and provider initialization.

Dec 16, 2025

Improves Bedrock integration by passing S3 URLs directly to the API and updates documentation. It also fixes instruction insertion for models that don't natively support instructions.

Dec 16, 2025

Adds tool timeout support and allows Temporal Agents to be registered to workflows. It also replaces OpenTelemetry events with logs and fixes UIAdapter typing.

Dec 13, 2025

Restores prompt caching support for AWS Bedrock and adds new methods for schema generation and provider URLs. It also adds GPT-5.2 support and updates OpenAI library.

Dec 12, 2025

A hotfix release that restores compatibility with OpenAI v1 and temporarily reverts prompt caching options due to compatibility issues.

Dec 11, 2025

This release adds prompt caching options to OpenAIChatModelSettings, introduces support for multi-modal output in LLMJudge, adds CerebrasModel support, and fixes several bugs including GraphBuildingError and BedrockConverseModel issues.

Dec 11, 2025

This release adds support for aspect ratio in Gemini image generation, passes container_id back to Anthropic API, upgrades temporalio to 1.20.0, and fixes several bugs related to OpenRouter responses and GoogleModel thinking signatures.

Dec 10, 2025