LLM 0.32a2 Release: OpenAI Models Switch to /v1/responses Endpoint
Patch your LLM usage to call /v1/responses for reasoning‑capable OpenAI models, or use –R to hide reasoning.
Patch your LLM usage to call /v1/responses for reasoning‑capable OpenAI models, or use –R to hide reasoning.
Summary
The LLM library released version 0.32a2 on GitHub, adding several useful features to its alpha. The most significant change is that reasoning‑capable OpenAI models now use the /v1/responses endpoint instead of the traditional /v1/chat/completions. This switch unlocks interleaved reasoning across tool calls for GPT‑5‑class models, allowing developers to see the step‑by‑step logic in the API response. The new endpoint returns summarized reasoning tokens, which are rendered in a distinct color in the CLI output.
Users can still suppress the reasoning display by passing the –R or --hide-reasoning flags. The change is documented in pull request #1435 and is reflected in the updated LLM codebase. The release also includes minor bug fixes and documentation updates, but the endpoint shift is the primary feature that developers need to adapt to. The LLM team encourages testing the new endpoint with existing OpenAI integrations to ensure compatibility.
Key changes
- OpenAI reasoning‑capable models now use /v1/responses instead of /v1/chat/completions
- Interleaved reasoning across tool calls is enabled for GPT‑5‑class models
- Summarized reasoning tokens are displayed in a distinct color
- The –R or --hide-reasoning flags suppress reasoning output
- Pull request #1435 documents the change and updates the code