2026-08-17
Pydantic AI v2.31.1 adds Bedrock and Gemini compatibility fixes
Pydantic AI v2.31.1 lists two model-compatibility bug fixes: it denies native structured output for Claude Sonnet 5 and Fable 5 on Bedrock, and it falls back to `thinking_level='LOW'` when Gemini models reject `MINIMAL`.
Pydantic AI v2.31.1 documents two compatibility-oriented bug fixes: one changes native structured-output handling for specified Bedrock models, while the other adds a Gemini fallback when the `MINIMAL` thinking level is not accepted.
01
What we know now
- 01
[1] Pydantic AI official GitHub release record for v2.31.1, titled v2.31.1 (2026-08-17): https://github.com/pydantic/pydantic-ai/releases/tag/v2.31.1.
- 02
The official release notes list denial of native structured output for Claude Sonnet 5 and Fable 5 on Bedrock.
- 03
The official release notes list a fallback to `thinking_level='LOW'` for Gemini models that reject `MINIMAL`.
02
Why this matters for Armenia
The update is relevant to Armenian developers and teams using Python-based AI tooling, as it documents model-specific compatibility behavior they may encounter when upgrading.
03
Bedrock models
Native structured output is denied for the named models.Gemini models
When `MINIMAL` is rejected, the setting falls back.Release scope
The release record lists these items as bug fixes.Summary of the two model-compatibility changes listed in the official v2.31.1 release notes.
04
Bedrock structured-output handling
The v2.31.1 notes identify native structured output as unsupported for Claude Sonnet 5 and Fable 5 when used on Bedrock. Developers relying on that capability should validate their output-handling path after updating, particularly if their application assumes native schema-constrained responses.
- The official v2.31.1 release record lists a bug fix that denies native structured output for Claude Sonnet 5 and Fable 5 on Bedrock.
- This is a targeted compatibility change for those specifically named model and platform combinations.
05
Gemini thinking-level fallback
The release also addresses Gemini compatibility around the thinking-level setting. According to the official notes, requests using `MINIMAL` receive a fallback to `LOW` when the model rejects the original value. This can help preserve operation across Gemini models with differing accepted thinking-level options.
- If a Gemini model rejects `MINIMAL`, Pydantic AI falls back to `thinking_level='LOW'`.
- The stated behavior is a fallback, rather than a claim that every Gemini model accepts the same setting.
06
Scope of the update
Pydantic AI presents v2.31.1 as a bug-fix release with the two compatibility changes described above. The release notes support evaluating this update chiefly where applications use the affected Bedrock structured-output route or Gemini thinking-level configuration.
- The release is titled v2.31.1 and is dated 2026-08-17 in the official record.
- The linked comparison covers changes from v2.31.0 to v2.31.1.
07
Practical update steps
For teams using the affected model paths, the official notes support a focused upgrade review.
- 01
Review integrations that request native structured output through Bedrock for Claude Sonnet 5 or Fable 5.
- 02
Test Gemini workflows that use a `MINIMAL` thinking level and confirm the fallback behavior is acceptable.
- 03
Compare v2.31.1 with v2.31.0 using the release changelog and run existing integration tests after upgrading.
08
Limits of this edition
This summary reflects the official release notes and does not independently test the changes.
The release notes do not provide broader implementation detail, affected configurations beyond those named, or performance impact.
SRC
Source desk
Direct links to the material behind this selection. Seeing the source matters as much as reading the synthesis.