Pydantic AI 2.68.0 released
A feature release adding conversation storage, improved model fallback handling, and durable delegation support for sub-agents.
What changed in 2.68.0
- Load and save UI adapter conversations through ConversationStore interface.
- Fall back to next model in ModelFallback when a response hits max_tokens limit.
- Request user approvals and answers inside runs with RequestInput, moving AskUser to deferred pipeline.
- Support running SubAgents delegations as Temporal and DBOS child workflows.
- Add RefineOutput harness capability and GoogleDecisionsModel backend for Gemini Decisions API.
Check before you update
- Sub-agent task-control events renamed with Event suffix and past-participle names; old names deprecated.
- Harness identifiers expanded: num_results renamed to max_results, max_duration_secs to max_duration_seconds.
- CapabilityEvent field names changed to *_count, *_name, duration_ms, previous_item; old names deprecated.
Summarised by AgentGid from Pydantic's official release notes on GitHub, as reported there; read the original before relying on it. Price, status and release dates come from our daily data.
Pydantic AI at a glance
| What it is | A Python agent framework from the Pydantic team, built around type-safe structured outputs, dependency injection and model-agnostic tool calling. |
|---|---|
| Latest version | 2.71.0, Nov 19, 2026. This page is about an earlier release. |
| Release pace | 15 stable releases in the last 30 days; 2.68.0 came 2 days after the previous one |
| Price | Free (OSS). Pydantic AI is open source and free to use; you pay only for the models and infrastructure it runs on. (checked Nov 4, 2026) |
| Downloads, last 7 days | 1.6M up 1.1% |
| Gid Score | 53 #15 in Agent frameworks and SDKs |