Skip to content

Bug: Incompatibility between GPT-5.4 "Responses API" and UI-forced reasoning_effort parameter #2826

@EliYagel

Description

@EliYagel

Issue you'd like to raise.

When using the OpenAI Responses API provider configuration with gpt-5.4-mini, the LangSmith UI forces a Reasoning Effort selection (Default/Low/Medium/High) with no option to select "None" or "Disabled."
This causes two critical failures:

  • API Crash: The UI passes the parameter as a top-level keyword argument reasoning_effort. The newer OpenAI Responses API endpoint does not accept this; it requires a nested reasoning: {"effort": "..."} object. This results in: TypeError: AsyncResponses.create() got an unexpected keyword argument 'reasoning_effort'.
  • Schema Conflict: Because the UI hard-codes reasoning_effort, users cannot manually add the required reasoning: {"summary": "auto"} block into the "Extra Parameters" JSON. The UI validator prevents saving the JSON due to a duplicate/conflicting key error, making it impossible to enable Reasoning Summaries.

Suggestion:

  • Update UI Dropdown Logic: When the "Responses" API is selected, the Reasoning Effort dropdown should map to the nested reasoning.effort path instead of the top-level reasoning_effort key.
  • Add "Summary" Toggle: Add a UI toggle for Reasoning Summary (auto/none) so users don't have to manually inject it via JSON.
  • Add "None/Override" Option: Allow the Reasoning Effort dropdown to be set to "None" or "Manual/JSON Override" to prevent the UI from injecting hard-coded parameters that conflict with user-defined JSON in the "Extra Parameters" field.

Metadata

Metadata

Assignees

Labels

bugSomething isn't workingsdk

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions