Sam Altman Says OpenAI “Screwed Up” GPT-5.2 Writing Quality

Sam Altman mentioned OpenAI “screwed up” GPT-5.2’s writing high quality throughout a developer town hall Monday night.

When requested about person suggestions that GPT-5.2 produces writing that’s “unwieldy” and “laborious to learn” in comparison with GPT-4.5, Altman was blunt.

He mentioned:

“I feel we simply screwed that up. We are going to make future variations of GPT 5.x hopefully significantly better at writing than 4.5 was.”

Altman defined that OpenAI made a deliberate option to focus GPT-5.2’s improvement on technical capabilities:

“We did resolve, and I feel for good cause, to place most of our effort in 5.2 into making it tremendous good at intelligence, reasoning, coding, engineering, that form of factor. And we’ve restricted bandwidth right here, and typically we deal with one factor and neglect one other.”

How OpenAI Positioned Every Mannequin

The distinction between GPT-4.5 and GPT-5.2 exhibits the place OpenAI centered its sources.

When OpenAI introduced GPT-4.5 in February 2025, the corporate emphasised pure interplay and writing. OpenAI mentioned interacting with GPT-4.5 “feels extra pure” and referred to as it “helpful for duties like bettering writing.”

GPT-5.2’s announcement took a special route. OpenAI positioned it as essentially the most succesful mannequin sequence but for skilled information work, with enhancements in creating spreadsheets, constructing shows, writing code, and dealing with advanced, multi-step initiatives.

The discharge submit spotlights spreadsheets, shows, instrument use, and coding. Writing seems extra briefly, with technical writing famous as an enchancment for GPT-5.2 Immediate. However Altman’s feedback recommend the general writing expertise nonetheless fell brief for customers evaluating it to GPT-4.5.

Why This Issues

We’ve coated the iterative adjustments to ChatGPT since GPT-5 launched in August, together with updates to heat and tone and the GPT-5.1 instruction-following enhancements. OpenAI frequently adjusts mannequin conduct based mostly on person suggestions, and regressions in a single space whereas bettering one other aren’t new.

What’s uncommon is listening to Altman acknowledge a tradeoff this instantly. For anybody utilizing ChatGPT output in client-facing work, drafts, or polished writing, this explains why outputs might have modified. Mannequin upgrades don’t assure enchancment throughout each functionality.

If you happen to depend on ChatGPT for writing, deal with mannequin updates like another dependency change. Re-test your prompts when defaults change, and hold a fallback if output high quality issues in your workflow.

Wanting Forward

Altman mentioned he believes “the long run is usually going to be about excellent normal goal fashions” and that even coding-focused fashions ought to “write nicely, too.”

No timeline was given for when GPT-5.x writing enhancements will ship. OpenAI usually iterates on mannequin conduct by means of level releases, so adjustments might arrive progressively fairly than in a single replace.

Hear Altman’s full assertion within the video beneath:

Featured Picture: FotoField/Shutterstock