What it is
OpenAI’s official prompt guidance page with model-specific prompting advice (including GPT‑5.5) and practical tips for migrating older prompt stacks.
Gabriel’s notes
An official OpenAI API documentation page that collects model-specific prompting guidance (e.g., GPT‑5.5, GPT‑5.4, GPT‑5.3, GPT‑5.x, GPT‑4.1) and highlights what to change when migrating older prompt stacks to newer models—especially when you’re building tool-using workflows in the Responses API.
Quick take: This is the closest thing OpenAI has to a “stop guessing, start here” page for prompts. When I’m updating an existing system prompt, I check this first—because stale prompt lore ages like milk.
My original note still stands with one small lawyerly tweak: it’s the most up-to-date prompting guidance from OpenAI for a set of major model families (including GPT‑5.5 at the time I saved this on May 12, 2026). The claim that it covers every OpenAI model is Unknown / not confirmed—it’s clearly broader than a single model, but it’s organized around selected families/tabs rather than an exhaustive catalog.
I saved this under Prompting because this page is where “prompt engineering” stops being vibes and starts being versioned migration guidance.
Good fit if you want to:
- Shorten bloated, process-heavy prompt stacks into cleaner, outcome-first instructions.
- Migrate older prompts to GPT‑5.5 without accidentally making your assistant weirdly robotic.
- Design more reliable agent/tool behavior (system prompt structure, phases, replay, etc.).
- Make personality/tone constraints explicit for customer-facing assistants (instead of hoping for the best).
- Debug failures by iterating prompts alongside evals and real traces (not just “prompt harder”).
Pricing snapshot (auto-enriched)
Reading the docs is free; using the API is usage-based. OpenAI’s model docs list gpt-5.5 at $5 per 1M input tokens and $30 per 1M output tokens (check current numbers before you ship), and the API pricing page lists token pricing across many other models as well.
Work-use / compliance snapshot (auto-enriched)
Policy: Your application and users still need to comply with OpenAI’s Usage Policies (and relevant terms). Usage Policies are dated and change over time—treat them like living requirements, not a one-time checkbox.
Data handling (API): OpenAI’s “Your data” docs say API data is not used to train/improve models by default (unless you explicitly opt in). They also describe default retention for abuse monitoring logs (commonly 30 days for key endpoints) and options like Modified Abuse Monitoring / Zero Data Retention (eligibility/approval required), plus some model-specific caveats. If you handle sensitive data, read this page end-to-end and align it with your internal policies and any customer contracts.
Alternatives (auto-enriched)
- OpenAI: Best practices for prompt engineering (Help Center) — shorter, more general “prompting 101” guidance; great for onboarding teammates.
- GPT‑5 for Coding (cheatsheet PDF) — more tactical if your primary workload is coding/agentic code edits rather than general assistant behavior.
Before you adopt it:
- Version your prompts like code (git + changelog) and tie changes to eval results, not feelings.
- When migrating, remove “legacy scaffolding” first—then re-add only what fixes observed failures.
- Write down an explicit output contract (format, constraints, success criteria) before you tune tone/personality.
Sources
- https://developers.openai.com/api/docs/guides/prompt-guidance?model=gpt-5.5
- https://developers.openai.com/api/docs/models
- https://platform.openai.com/docs/pricing/
- https://developers.openai.com/api/docs/guides/your-data
- https://platform.openai.com/docs/usage-policies/usage-policies