Skip to content

[BUG]: Playground allows invalid temperature + top_p combination for all Claude 4.x models #11278

@jash0803

Description

@jash0803

Where do you use Phoenix

Self-hosted

What version of Phoenix are you using?

12.34.0

What happened?

In the Phoenix Playground, all Claude 4.x models allow configuring both temperature and top_p simultaneously via the UI.

This results in a consistent runtime failure when executing the prompt:

invalid_request_error: 'temperature' and 'top_p' cannot both be specified for this model. Please use only one.

This is reproducible across every Claude 4.x variant, including:

  • claude-opus-4-5
  • claude-opus-4-5-20251101
  • claude-sonnet-4-5
  • claude-sonnet-4-5-20250929
  • claude-haiku-4-5
  • claude-haiku-4-5-20251001
  • claude-opus-4-1
  • claude-opus-4-1-20250805

What did you expect to happen?

The Playground UI should prevent invalid parameter combinations for Claude 4.x models by enforcing that only one of temperature or top_p is set.

Acceptable solutions would be:

  • Disable one control when the other is set
  • Automatically omit one parameter from the request
  • Surface a UI validation error before execution

Why this is critical

This is a breaking change introduced when migrating from Claude 3.x to Claude 4.x, where:

  • Claude 3.x allowed both parameters
  • Claude 4.x explicitly rejects requests containing both

Because the Playground still exposes both controls without validation, every run fails by default unless the user manually discovers and fixes the issue.

Impact:

  • Playground is effectively broken for all Claude 4.x models
  • Blocks prompt iteration and experimentation
  • Produces confusing errors that look like platform failures
  • Affects new users and existing users migrating models

Additional information

Image

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingc/playgroundprompt playground and llm provider support

    Type

    No type

    Projects

    Status

    📘 Todo

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions