Pydantic AI v2.68.0
Pydantic AI v2.68.0 released with conversation store, streamed cost for Google and Bedrock models, durable sub-agent delegations, and RefineOutput harness capability.
The Deploy Log | Companies | Free, no signup
Every row from Pydantic in The Deploy Log: 19 rows, newest first, each linking the publisher's own page. The newest shipped are Pydantic AI v2.68.0 and Pydantic AI v2.67.0. The first is dated and the newest . By lane: Tools and products 19. They fall in 3 months. The tables below count them by lane, month and platform, and date them.
| Lane | Rows |
|---|---|
| Tools and products | 19 |
| Figure | Value |
|---|---|
| First row dated | |
| Newest row dated | |
| Median days between release days, publisher's dates only | 3 |
| Platform | Rows |
|---|---|
| Amazon Bedrock | 2 |
Pydantic AI v2.68.0 released with conversation store, streamed cost for Google and Bedrock models, durable sub-agent delegations, and RefineOutput harness capability.
Pydantic AI v2.67.0 adds durable execution subagents, instruction file loading, and provider settings changes.
Pydantic AI v1.107.10 raises the remaining web_fetch timing bounds to 60 seconds.
Added 12h prompt cache retention support for GPT-5.6 and later on OpenAI Responses API.
pydantic-ai v2.63.0 defaults CodeMode to dynamic_catalog=True and changes Gemini Live reconnect and Harness LLMReminder defaults.
Pydantic AI v2.62.0 adds mid-conversation thinking effort changes and new model error categories.
pydantic-ai v2.61.0 released with thinking override warnings and Cohere tool return handling.
Pydantic AI v2.60.0 released with model error categories, ModelSelectionContext fields, Bedrock maxTokens default, and deprecations.
Fixed GHSA-4whf-38hw-pxxh where attacker-controlled HTML in web_fetch could delay unrelated work.
Fixed GHSA-4whf-38hw-pxxh where attacker-controlled HTML in web_fetch could delay unrelated work.
Fixed MCP request telemetry exposure of server URLs, tokens, and headers in OpenTelemetry traces.
Fixed MCP request telemetry exposure of server URLs, tokens, and headers in OpenTelemetry traces.
Pydantic AI v2.57.0 adds MetaProvider for Meta Muse models and updates realtime session handling.
Pydantic AI v2.56.0 adds Skills layering, ZaiProvider base_url, PGTaskDurability, and Render Workflows integration.
Pydantic AI v2.55.0 requires Python 3.11 or newer and adds durable execution for ExaSearch, YouSearch, and LocalStack tool requests.
Released pydantic-ai v2.54.0 with fixes for JSON schema, image generation, and durable execution.
Fixes a high-severity concurrency slot leak in ConcurrencyLimitedModel.
Adds DecisionModel and exposes OpenAI service_tier in provider details.
Adds OpenAI GPT-Live support with OpenAILiveModel and exposes context_window_used.
One line every time AI actually ships, around the clock, free on Threads and Bluesky. Every one is in The Deploy Log with its source.