What's Changed

• Improve validation error retry message by @dmontagu in #3193
• OpenAI gpt-5-chat does not support (encrypted) reasoning by @DouweM in #3332
• Complete thinking.md documentation with AWS Bedrock examples by @daniilr in #3328
• Fix graph execution bug with multiple joins downstream of same fork by @dmontagu in #3337
• Skip installing outlines dependencies mlx, vllm, torch on Intel Macs by @pcustic in #3312
• Let additional instructions be provided at agent.run time. by @baek54321 in #3309
• Add partial_output to RunContext provided to output validators by @petersli in #3286

New Contributors

• @daniilr made their first contribution in #3328
• @pcustic made their first contribution in #3312
• @baek54321 made their first contribution in #3309
• @petersli made their first contribution in #3286

Full Changelog: v1.10.0...v1.11.0

What's Changed

• Fix MCPServer error handling with Temporal by @wreed4 in #3299
• Update directory path in ag-ui.md example by @sabman in #3310
• Implement OpenAIResponsesModel.base_url property by @chasefarmer-pixee in #3281
• Fix types to let OpenRouterProvider be created with only http_client by @xcpky in #3308
• Add Agent.run_stream_sync method and sync convenience methods on StreamedRunResult by @ajac-zero in #3146
• Fix typevar variance for agent deps by @dmontagu in #3319
• Add support for detecting and handling application/msword files. by @DenysMoskalenko in #3318
• Ensure AG-UI ToolCallStartEvent doesn't use a parent_message_id from a previous request/response by @DouweM in #3325

New Contributors

• @wreed4 made their first contribution in #3299
• @sabman made their first contribution in #3310
• @chasefarmer-pixee made their first contribution in #3281
• @xcpky made their first contribution in #3308
• @ajac-zero made their first contribution in #3146
• @DenysMoskalenko made their first contribution in #3318

Full Changelog: v1.9.1...v1.10.0

What's Changed

• Support AsyncAnthropicVertex as AnthropicProvider.anthropic_client by @DouweM in #3292
• Set AG-UI frontend state directly on provided deps so it can be read from on_complete handler by @DouweM in #3297
• Retry instead of error when Google response is empty with MALFORMED_FUNCTION_CALL or other recoverable finish reason by @ArneZsng in #3300
• Add Version Policy to docs by @DouweM in #3301

Full Changelog: v1.9.0...v1.9.1

What's Changed

• Support Vercel AI Data Stream Protocol by @samuelcolvin in #2923
• Fix docs custom retry logic example by @yf-yang in #3276

New Contributors

• @yf-yang made their first contribution in #3276

Full Changelog: v1.8.0...v1.9.0

What's Changed

• Add experiment metadata by @dmontagu in #3263
• Respect openai_supports_tool_choice_required model profile setting in OpenAIResponsesModel by @pamelafox in #3272
• Use latest OpenAI, Google, Anthropic models in all examples by @DouweM in #3278
• Fix agent name inference when using run_stream_events by @DouweM in #3279

New Contributors

• @pamelafox made their first contribution in #3272

Full Changelog: v1.7.0...v1.8.0

What's Changed

• Add OutlinesModel to run local models using Transformers, Llama.cpp, MLXLM, SGLang and vLLM via Outlines by @RobinPicard in #2623
• Fix pydantic-graph importing pydantic-ai by @dmontagu in #3265

New Contributors

• @RobinPicard made their first contribution in #2623

Full Changelog: v1.6.0...v1.7.0

What's Changed

• Add FastMCPToolset by @strawgate in #2784
• Add OpenAIModelProfile.openai_responses_requires_function_call_status_none flag to satisfy vLLM Responses API by @DouweM in #3246
• Sanitize auto-generated output tool name to support generic types by @lionpeloux in #2979
• Ensure ToolCallPart.args resulting from TestModel(custom_output_args=...) is always a dict by @DouweM in #3254

Full Changelog: v1.5.0...v1.6.0

What's Changed

• Introduce new graph API in beta by @dmontagu in #2982
• Preformat run graph/node span names for other OTel backends by @alexmojaki in #3244
• Ensure that google-genai doesn't close httpx client provided by Pydantic AI or user by @DouweM in #3243

Full Changelog: v1.4.0...v1.5.0

What's Changed

• Support OpenAI and Anthropic native MCP support via MCPServerTool builtin tool by @Artui in #3101, #3239
• Use correct agent's instructions when telling model to retry output by @dsfaccini in #3209
• Raise clear error when any Google content filter is hit resulting in empty response by @DouweM in #3236
• Expand docs for pydantic-evals by @dmontagu in #3213

New Contributors

• @Artui made their first contribution in #3101

Full Changelog: v1.3.0...v1.4.0

What's Changed

• Raise IncompleteToolCall when token limit is reached during generation of tool call by @erhuve in #3137
• Include evals report averages in span attributes by @cetra3 in #3053
• Make AbstractBuiltinTool serializable and work with durable execution by @DouweM in #3176
• Ignore empty text deltas when streaming gpt-oss on Bedrock by @DouweM in #3215
• Ignore empty text deltas when streaming gpt-oss via Ollama by @DouweM in #3216
• docs: Replace deprecated Cohere model 'command' with 'command-r7b-12-… by @sijanonly in #3201
• Include all usage fields in OTel attributes by @DouweM in #3221
• Correct Prefect .serve example by @desertaxle in #3192
• Ensure toolset spans (e.g. MCP sampling) are nested under agent run span by @DouweM in #3224
• Update genai-prices, fix default api_flavor, fixes anthropic usage extraction by @alexmojaki in #3226
• feat(gateway): support AWS Bedrock by @Kludex in #3203
• Add OVHcloud AI Endpoints provider by @eliasto in #3188
• Use gateway/<upstream_provider>: as provider name prefix for Gateway by @DouweM in #3229
• Add http_client option to GoogleProvider, support api_key for Vertex AI, and use Pydantic AI's cached httpx client by default by @DouweM in #3217

New Contributors

• @erhuve made their first contribution in #3137
• @cetra3 made their first contribution in #3053
• @sijanonly made their first contribution in #3201
• @eliasto made their first contribution in #3188

Full Changelog: v1.2.0...v1.3.0