GPT-5.5 prompting guide
TLDR
- OpenAI released a GPT-5.5 prompting guide recommending builders start from a fresh prompt baseline, not migrate existing GPT-5.2 or GPT-5.4 prompts.
Key Takeaways
- OpenAI explicitly warns against treating GPT-5.5 as a drop-in replacement; tune reasoning effort, verbosity, tool descriptions, and output format from scratch.
- For multi-step agentic tasks, send a short one-to-two sentence user-visible acknowledgment before any tool calls to prevent the interface from feeling frozen.
-
The Codex app ships an
openai-docsskill; running$openai-docs migrate this project to gpt-5.5triggers a guided automated migration using the official upgrade guide. - The upgrade guide includes instructions on rewriting prompts to fit GPT-5.5’s behavior, not just swapping model IDs.
- Simon Willison notes OpenAI’s own Codex app already uses the early-acknowledgment pattern, and observes it meaningfully improves perceived responsiveness.
Why It Matters
- Any production system on GPT-5.2 or GPT-5.4 that port prompts without re-tuning is likely to get degraded or unpredictable outputs from GPT-5.5.
- The early-acknowledgment pattern is a low-cost UX fix for agentic apps where latency before first output causes users to assume failure.
- OpenAI embedding migration advice directly into a Codex skill sets a precedent for model-version transitions as tooling problems, not just documentation problems.
Simon Willison, Simon Willison’s Weblog · 2026-04-25 · Read the original
