[dependabot]

Updates the requirements on pydantic-ai-slim to permit the latest version.

Release notes

Sourced from pydantic-ai-slim's releases.

v2.64.0 (2026-10-30)

What's Changed

⚠️ Compatibility Notes

• Raise an error status a retrying transport gave up on as ModelHTTPError, so fallback moves on by @​DouweM in pydantic/pydantic-ai#11934
• Send stored response, item, conversation and container IDs back only to the endpoint that created them, via Model.accepts_server_state by @​DouweM in pydantic/pydantic-ai#11774
• Make json-repair an optional dependency, needed only by Coder and RepairToolArguments by @​mpfaffenberger in pydantic/pydantic-ai#11768
• Turn off strict tool schemas for Coder's tools by default, with Coder(strict=True) to opt in by @​mpfaffenberger in pydantic/pydantic-ai#11743
• Surface every nested RepoContext instruction file between a traversed path and the working directory, without repeating the root file by @​DouweM in pydantic/pydantic-ai#11687
• Add output_speech_timestamps to report word timing of realtime speech, on Azure AI Voice Live by @​DouweM in pydantic/pydantic-ai#11685
• Make the ModelFallback capability constructible from agent specs by @​DouweM in pydantic/pydantic-ai#11679
• Add a first_event_timeout model setting so a stream that never starts can fall back by @​DouweM in pydantic/pydantic-ai#11582
• Dispatch OnlineEvaluation evaluators once per durable run instead of on every replay by @​DouweM in pydantic/pydantic-ai#11577
• Show the harness Planning plan in the plan tools' results instead of a per-request reminder copy by @​DouweM in pydantic/pydantic-ai#11521
• Add agent artifact security scanning for Memory and Skills with ArtifactScanner by @​mpfaffenberger in pydantic/pydantic-ai#11470
• Fix AnthropicModel.count_tokens rejecting all-deferred tool requests and undercounting server tools on the Anthropic API by @​dsfaccini in pydantic/pydantic-ai#11343
• Accept plain functions as evaluators in pydantic_evals by @​dmontagu in pydantic/pydantic-ai#10292
• Remove unresolvable RealtimeEvent from pydantic_ai.agent.__all__ by @​adtyavrdhn in pydantic/pydantic-ai#10189
• Add RunContext.load_capability to load on-demand capabilities from code by @​adtyavrdhn in pydantic/pydantic-ai#10090
• Add a ModelFallback capability, a per-attempt prepare_model_request hook, and RetryModelRequest by @​DouweM in pydantic/pydantic-ai#9348

🚀 Features

• Let @agent.system_prompt(dynamic=True) append its changes with on_change='append' to keep the cached prompt by @​DouweM in pydantic/pydantic-ai#11445
• Pair DetectLoops tool results with the response that made the calls by @​mpfaffenberger in pydantic/pydantic-ai#11551
• Add BoxdSandbox: a workspace that runs Coder, Shell, FileSystem and every other harness capability in a boxd machine by @​MichielMAnalytics in pydantic/pydantic-ai#11367
• Add a ModelFallback capability, a per-attempt prepare_model_request hook, and RetryModelRequest by @​DouweM in pydantic/pydantic-ai#9348
• Add Harness Reminder(scope='conversation') for reminders that stay in the message history by @​DouweM in pydantic/pydantic-ai#11642
• Add output_speech_timestamps to report word timing of realtime speech, on Azure AI Voice Live by @​DouweM in pydantic/pydantic-ai#11685
• Cache the schema of ModelRequestContext durable capability operations across agent construction by @​DouweM in pydantic/pydantic-ai#11776
• Make the ModelFallback capability constructible from agent specs by @​DouweM in pydantic/pydantic-ai#11679
• Report word timestamps of ElevenLabs Agents speech with output_speech_timestamps by @​DouweM in pydantic/pydantic-ai#11784
• Record CI provenance on Pydantic Evals experiment spans with OpenTelemetry vcs.* and cicd.* attributes by @​dmontagu in pydantic/pydantic-ai#10294
• Accept plain functions as evaluators in pydantic_evals by @​dmontagu in pydantic/pydantic-ai#10292
• Add logfire.scorer.score_type and logfire.scorer.run_id to online evaluation result events by @​dmontagu in pydantic/pydantic-ai#10293
• Add the OpenTelemetry test.suite.name and test.case.name attributes to Pydantic Evals spans by @​dmontagu in pydantic/pydantic-ai#11538
• Turn off strict tool schemas for Coder's tools by default, with Coder(strict=True) to opt in by @​mpfaffenberger in pydantic/pydantic-ai#11743
• Make json-repair an optional dependency, needed only by Coder and RepairToolArguments by @​mpfaffenberger in pydantic/pydantic-ai#11768
• Teach Coder to read what the task names, check stated requirements, and delegate only when asked by @​mpfaffenberger in pydantic/pydantic-ai#11811
• Let AskUser answerers receive RunContext by @​dsfaccini in pydantic/pydantic-ai#11662
• Add agent artifact security scanning for Memory and Skills with ArtifactScanner by @​mpfaffenberger in pydantic/pydantic-ai#11470
• Add ModelResponse.model_id recording the model the agent sent each request to by @​DouweM in pydantic/pydantic-ai#11807
• Save function evaluators to Dataset files under the function's name by @​dmontagu in pydantic/pydantic-ai#10425
• State how nested instruction files relate in RepoContext instructions when nested_traversal is on by @​DouweM in pydantic/pydantic-ai#11698
• Add a first_event_timeout model setting so a stream that never starts can fall back by @​DouweM in pydantic/pydantic-ai#11582
• Add sticky to ModelFallback to keep a conversation on the fallback model that answered by @​DouweM in pydantic/pydantic-ai#11330
• Add RunContext.load_capability to load on-demand capabilities from code by @​adtyavrdhn in pydantic/pydantic-ai#10090

🐛 Bug Fixes

• Preserve nested SpanTree attributes during JSON round-trips and mark omitted Google embedding and ElevenLabs usage as unmeasured to avoid false zero automatic costs by @​dsfaccini in pydantic/pydantic-ai#11912
• Fall back from Native Output on Bedrock when the installed botocore predates outputConfig by @​DouweM in pydantic/pydantic-ai#11694
• Close the IPythonKernel client's zmq context on shutdown with jupyter-client before 8.7 by @​DouweM in pydantic/pydantic-ai#11716

... (truncated)

Changelog

Sourced from pydantic-ai-slim's changelog.

---

description: "Pydantic AI upgrade guide and changelog: the breaking changes in each release, how to install V2, and the recommended path to migrate your code from V1."

Upgrade Guide

In September 2025, Pydantic AI reached V1 and committed to API stability: no changes that break your code until V2. V2 is now available, collecting the breaking and behavior changes that stability guarantee didn't allow. This guide is the canonical place to learn what's in V2, how to install it, and how to upgrade; for the guarantees behind these version numbers, see the Version Policy.

Deprecations

Functionality deprecated in a V2 release keeps working until V3, and using it raises a PydanticAIDeprecationWarning naming its replacement. Resolving these as they appear is the smooth path to the next major, the same way resolving V1's warnings was the smooth path to this one.

Deprecated in V2, to be removed in V3

• Calling a [RunUsage][pydantic_ai.usage.RunUsage], as in V1's result.usage(), returns that same usage and warns. usage became a property in V2, and the call used to raise TypeError: 'RunUsage' object is not callable, which code that wraps usage accounting in except Exception turned into silently zero usage. Drop the parentheses: result.usage.
• [RetryPromptPart][pydantic_ai.messages.RetryPromptPart] is deprecated in favor of the two parts it conflated: a retry that answers a tool call is that call's own [ToolReturnPart][pydantic_ai.messages.ToolReturnPart] with outcome='retried', and one that answers no tool call is a [RetryFeedbackPart][pydantic_ai.messages.RetryFeedbackPart]. The old part reached the model as bare user text in the second case, which the model couldn't tell from something a person wrote (#6404, #7799). Pydantic AI no longer emits it, and constructing one warns.
  • What keeps working: a 'retry-prompt' part in a serialized history loads as whichever of the two it always meant, and an instance your own code holds, including one handed back through [DeferredToolResults][pydantic_ai.tools.DeferredToolResults], is translated the same way before it reaches a model. The direct helpers send it in that new shape too. Calling [Model.request()][pydantic_ai.models.Model.request] or [Model.count_tokens()][pydantic_ai.models.Model.count_tokens] yourself skips that translation but still works: a tool-bound one goes out as that call's retried result, and a tool-less one as the user text it always rendered as. [sanitize_messages][pydantic_ai.messages.sanitize_messages] keeps client-returned retry feedback: it translates a tool-less one into the RetryFeedbackPart it always meant, keeps a 'validation_error' or 'model_retry' part as sent (it reaches the model fenced, in the user voice), and demotes a 'no_output' part, the one cause rendered in the system voice, to that same fenced, attributed user text, since a client can forge it; pass strip_system_prompts=False to keep a 'no_output' part as sent.
  • What you need to change: isinstance(part, RetryPromptPart) no longer matches, so match the two parts above instead. Each records why the retry happened, and which attempt it is against which limit, in a [RetryInfo][pydantic_ai.messages.RetryInfo] on its retry field, which tool spans also carry as pydantic_ai.tool.retry.* attributes. [ToolRetryError.tool_retry][pydantic_ai.exceptions.ToolRetryError] holds one of those two parts, so read tool_name off it only after checking for a ToolReturnPart.
  • What the model now sees: a tool retry carries the serialized validation errors or the ModelRetry message, with no N validation errors: header and no Fix the errors and try again. suffix. One that answers no call arrives in the voice its cause calls for: validation errors and a ModelRetry message in the user voice, inside a <validation_errors> or <retry_feedback> fence behind one line saying the application, not the user, is speaking, and the framework's fixed request for usable output in the system voice. On providers without a native tool error channel, a retried tool call's validation errors go in the {"error": ...} wrapper as a JSON array rather than as a string holding one, and Anthropic and Bedrock get them as one text block. Client-returned retry feedback is kept by [sanitize_messages][pydantic_ai.messages.sanitize_messages] and the UI adapters rather than stripped, and a 'no_output' part among it goes out as fenced user text rather than in the system voice. Retries in a history stored before upgrading are rendered the new way too, since rendering happens when each request goes out, so a conversation's first request after upgrading misses the provider's prompt cache from its first retry onward, once. A [decision model][pydantic_ai.models.decision.DecisionModel] judges a retried tool call as that call's tool_return, where it had a retry entry. On ag-ui-protocol below 0.1.11, [AGUIAdapter.dump_messages][pydantic_ai.ui.ag_ui.AGUIAdapter.dump_messages] now also warns on a run that retried its own output. See Retries.
  • Vercel AI dumps keep a request's part order: [VercelAIAdapter.dump_messages][pydantic_ai.ui.vercel_ai.VercelAIAdapter.dump_messages] now writes a ModelRequest as one UIMessage per run of same-role parts, in the order they were authored, instead of putting every system part first. This applies with no retries at all: a request holding an enqueued SystemPromptPart after its UserPromptPart now dumps as a user message followed by a system message, and positional message ids shift with it. The request's metadata rides on its last user message, or its last message when it has none.

Breaking Changes

Here's a filtered list of the breaking changes for each version to help you upgrade Pydantic AI.

v2.0.0 (2026-06-23)

The stable V2.0 release. There are no new breaking or behavior changes since the betas; the full breaking-change list and recommended upgrade path are in the v2.0.0b1 entry below. Install it with:

uv add pydantic-ai

v2.0.0b7 (2026-06-10)

The seventh V2 beta, forked from v1.107.0. There are no new V2 breaking or behavior changes since v2.0.0b6 below — everything in that entry applies unchanged — but this beta picks up the latest V1 release on top, which adds Claude Fable 5 / Mythos 5 model support and OpenRouter prompt caching (CachePoint), plus known_model_names() and Anthropic fixes; see the v1.107.0 release notes for the full list.

Install it the same way, pinning the exact pre-release version:

pip/uv-add "pydantic-ai==2.0.0b7"

For the full breaking-change list and the recommended upgrade path, see the v2.0.0b1 entry below; the only difference is that the latest V1 to upgrade through first is now v1.107.0.

v2.0.0b6 (2026-06-04)

The sixth V2 beta, forked from v1.106.0. There are no new V2 breaking or behavior changes since v2.0.0b5 below — everything in that entry applies unchanged — but this beta picks up the latest V1 release on top, which adds api_host/timeout configuration and base seed mapping for the xAI provider, plus streaming and data-URI handling fixes; see the v1.106.0 release notes for the full list.

Install it the same way, pinning the exact pre-release version:

... (truncated)

Commits

• cde1d10 Check comparison page counts against their sources, and date and source each ...
• d691ad0 Surface every nested RepoContext instruction file between a traversed path ...
• f578674 Add RunContext.load_capability to load on-demand capabilities from code (#1...
• 5ce158b Raise an error status a retrying transport gave up on as ModelHTTPError, so...
• 3cc8fc1 Ignore max_retries in OpenAI, Anthropic and Gemini batch requests instead o...
• e64e7fa Document choosing one retry owner and compare recovery scopes in the retries ...
• 4723da2 Add sticky to ModelFallback to keep a conversation on the fallback model ...
• 6d206b0 Add a first_event_timeout model setting so a stream that never starts can f...
• 8b554a5 Say on the enterprise support page that it covers Pydantic AI Harness (#12020)
• 7e01966 Docs: link the related pydantic.dev blog posts from their docs pages (#12024)
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)

[dependabot]

Looks like pydantic-ai-slim is no longer updatable, so this is no longer needed.