Skip to content

Pi exposes unsupported GPT reasoning effort levels #505

Description

@dgokeeffe

Problem

ucode registers Databricks GPT models as custom openai-responses models in Pi. Without model-specific thinkingLevelMap metadata, Pi exposes its generic thinking levels for every GPT model. Selecting an unsupported level can cause the Responses API or Databricks AI Gateway to reject the request.

For example, GPT-6 Astra supports low, medium, high, xhigh, and max, but does not support none. GPT-5.4/5.5/5.6 and Pro variants have different supported subsets.

Proposed fix

Declare model-specific Pi reasoning metadata and map Pi's off choice to the Responses API's none value only for models that support it. Hide unsupported minimal, low, or off choices where necessary.

A focused implementation and regression tests are included in the accompanying pull request.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions