OpenAI Released a Free Prompt Rewriting Tool. Nobody Who Tried It Says It Works.

OpenAI says its new free tool will automatically rewrite legacy prompts for GPT-5.5, cleaning out the accumulated workarounds from the GPT-4 era. The tool exists. Whether it does what OpenAI claims is another question — and nobody who has looked closely at it says the answer is yes.

Simon Willison, an independent developer who reviewed the migration guide when it launched, described the output as boilerplate added on top of existing prompts rather than a clean rewrite, according to his blog. "The upgrade guide even includes light instructions on how to rewrite prompts," he noted. "Begin migration with the fresh baseline instead of carrying over every instruction from an older prompt stack." That is OpenAI's recommendation: run the tool, then manually evaluate and rewrite everything anyway.

OpenAI's formal position, issued in its prompting guide alongside the tool's release on April 23, is that legacy prompts are now actively harmful in GPT-5.5. Every instruction a developer added to paper over a quirk in an earlier model adds noise, narrows the model's search space, and produces overly mechanical answers. The tool automates the first pass at fixing that. The evaluation and judgment remain human.

That is the offer: pay twice the input cost per token, spend engineering time rewriting your prompts, and trust that the new model justifies both. The developer community on forums including DEV Community has responded with characteristic directness: "Every instruction you added to paper over a quirk in the previous model is now debt. Not using it yet? No urgency. gpt-5.4 isn't going anywhere."

OpenAI's Docs Skill, invoked inside its Codex coding agent via the command openai-docs migrate this project to gpt-5.5, rewrites prompts following the new guidance. It produces output. The question the tool does not answer, and that OpenAI's documentation does not resolve, is whether that output is a meaningful improvement or just a reformatted version of the accumulated workarounds.

What OpenAI says changed and why it matters

GPT-5.5, formally released to the API on April 23, is architecturally different enough from its predecessors that the old prompting playbook is counterproductive, according to the company's own guide. The model reasons more efficiently, so developers should stop micromanaging how it thinks and start describing what they want. The guiding metaphor shifts from detailed GPS to destination address.

The most concrete evidence of that shift is the recommended prompt structure itself. Role definitions, a brief statement of who the AI is and what it is doing, were dismissed by some researchers as unnecessary or counterproductive in recent model generations. OpenAI's GPT-5.5 guide puts them back at the top of its recommended schema, The Decoder reported. The prompting community's internal debate about role definitions is not over, but OpenAI has picked a side.

Also quietly changed: GPT-5.5's default reasoning effort is "medium," where GPT-5's default was "high." OpenAI recommends starting at medium and only pushing higher if evaluations show a measurable gain. According to the DEV Community post, "Higher effort can actually make outputs worse on tasks with weak stopping criteria." That is an admission, buried in the documentation, that more computation is not always better and that GPT-5.5 thinks more efficiently than its predecessor.

The pricing problem

GPT-5.5 costs $5 per million input tokens and $30 per million output tokens. GPT-5.4 costs $2.50 per million input tokens. For developers who spent months optimizing verbose, step-by-step prompts for GPT-4 and GPT-5.2-era models, the migration is not free in any meaningful sense. The prompt rewrite takes engineering time. The new model costs twice as much per token to run. The productivity gain from migrating is, as yet, unproven in production at scale.

GPT-5.5 has two variants: the standard version for agentic coding and multi-step tool workflows, and GPT-5.5-Pro for demanding multi-pass work where quality matters more than latency. The pro variant sits at the top of the pricing tier. For most developers using AI in production products, the cost-benefit equation for migrating from GPT-5.4 is not obvious.

Nobody has published benchmarks that resolve whether GPT-5.5's efficiency gains justify the 2x input cost for the general case. OpenAI's position is that GPT-5.5 is a new model family, not an incremental update, and that treating it as a drop-in replacement is the mistake. The company has released extensive documentation and an automated migration tool. Whether that is enough to drive adoption at the prices it is charging is the open question.

What OpenAI needs

The company needs enough developers to migrate that GPT-5.5 becomes the new standard, which justifies the higher pricing, which funds the next round of capability development. The free migration tool is the floor of that effort. The ceiling is developer trust that the upgrade is worth the cost and the engineering time.

That trust has not been established. The developer community's response, measured in forum posts, blog analyses like Willison's, and production evaluation threads, suggests GPT-5.4 will remain the default choice for many teams until someone demonstrates, with real codebases and real evals, that the migration pays off. OpenAI's tool removes one barrier. It does not remove the question.