AI Agent News/stories/pydantic-ai-v2-70-0-adds-streamguardrail-prunetoolresults-and-retryonmodelerrors.md

Frameworks

Pydantic AI v2.70.0 adds StreamGuardrail, PruneToolResults and RetryOnModelErrors

Pydantic AI released v2.70.0 on November 17, 2026, with new features for streamed output screening, tool results, sub-agents and model retries. The release also includes compatibility changes, including forwarding service_tier to xAI and recording the served tier.

Source: github.com/pydantic/pydantic-ai/releases/tag/v2.70.0

The release adds StreamGuardrail, which screens each streamed text, thinking and tool-call part before the caller receives it. A new PruneToolResults option lets the model replace a tool result it just received with a summary. RetryOnModelErrors retries the same model on transient model API errors.

Several changes affect agent setup. Skills can now read a loaded skill's bundled files with read_files=True and substitute ${SKILL_DIR} placeholders. PlaywrightBrowser can provision its CDP endpoint per run, and a parent agent can check on a running background delegation with check_task. SubAgent gains on_model_error, which chooses retry or return separately from on_failure.

The compatibility notes cover forwarding service_tier to xAI and raising ModelRequestRejectedError before OpenAIChatModel sends tools to GPT-6 models with reasoning on. OpenAI gpt-6.2-sol and gpt-6.2-astra model support was also added. Provider SDK extras are now capped below the next major version that has not been tested.

Why it matters if you build agents

Agents that stream output can now screen each part before it reaches the caller, and sub-agents can handle model errors with their own policy.

Developers who pass service_tier settings to xAI or send tools to GPT-6 models should check the compatibility notes, since behavior changed there.

Get stories like this every weekday

One short email with the day's agent tooling news.

Subscribe