llm 0.32a2

2026-05-12 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

The `llm` command-line tool has released version 0.32a2, introducing a significant update for OpenAI model interactions. This alpha release primarily shifts how reasoning-capable OpenAI models, specifically GPT-5 class models, access the API. Instead of the `/v1/chat/completions` endpoint, these models now utilize the `/v1/responses` endpoint. This change enables interleaved reasoning across tool calls, providing users with summarized reasoning tokens directly in the command-line output. These reasoning tokens are displayed in a distinct color, and users can opt to hide them using the `-R` or `--hide-reasoning` flags.

Key takeaway

For NLP Engineers integrating OpenAI's GPT-5 class models via the `llm` command-line tool, you should be aware of the API endpoint shift to `/v1/responses`. This change provides valuable interleaved reasoning tokens, which can aid in debugging and understanding model behavior. Consider leveraging these new visual cues to refine your prompt engineering and tool call strategies, or suppress them if your workflow demands cleaner output.

Key insights

OpenAI's GPT-5 class models now use `/v1/responses` for interleaved reasoning, enhancing tool call visibility.

Principles

• API endpoints dictate model capabilities
• Interleaved reasoning improves transparency

Method

The `llm` tool now routes reasoning-capable OpenAI models to the `/v1/responses` endpoint, allowing the display of summarized reasoning tokens in a distinct color, which can be toggled off.

In practice

• Use `llm` for OpenAI model access
• Observe reasoning tokens in output
• Hide reasoning with `-R` flag

Topics

• llm CLI
• OpenAI API
• GPT-5 Models
• Interleaved Reasoning
• Tool Calls

Code references

• simonw/llm

Best for: NLP Engineer, AI Engineer, Machine Learning Engineer, Prompt Engineer

Related on AIssential

• OpenAI Quietly Told You to Throw Away Your Prompt Stack — Modern LLMs perform best when given clear outcomes, not prescriptive step-by-step instructions.
• llm-openai-via-codex 0.1a0 — A new tool allows OpenAI model access using existing Codex CLI credentials.
• GLM 5.1 Agentic Coding Test with OpenCode | New Best Open Coding LLM? | Live Test — GM 5.1 excels in agentic coding and project understanding, outperforming some established models in specific benchmarks.
• llm-gemini 0.32a0 — The "llm-gemini" plugin now supports streaming reasoning tokens from Google's Gemini models via "llm>=0.32a0".
• Deep Dive into LLMs like ChatGPT — LLMs are trained in stages, from vast internet data to human-guided refinement, enabling complex, yet fallible, capabilities.
• OpenAI’s New Voice Models Can Reason, Translate, and Transcribe – All While You’re Still Talking — OpenAI's new voice models enable real-time reasoning, translation, and transcription for highly interactive AI agents.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.