Sam Altman admits OpenAI misjudged GPT-5.2’s writing quality in internal town hall

OpenAI CEO Sam Altman

• Sam Altman admits GPT-5.2 writing quality is worse than GPT-4.5
• GPT-5.2 focused on technical tasks over writing due to limited bandwidth.
• GPT-5.x updates will enhance writing quality using user feedback.

Did our AI summary help?

OpenAI CEO Sam Altman has openly admitted that the company made a misstep with GPT-5.2, particularly when it comes to writing quality. During a recent internal town hall meeting with employees, Altman acknowledged that users have complained the model’s writing feels unwieldy and harder to read when compared to GPT-4.5.

Altman did not attempt to soften the criticism. Responding to questions about user feedback, he said the company “just screwed that up” and added that future versions of GPT-5.x would aim to surpass GPT-4.5 in writing quality. He explained that OpenAI made a deliberate trade-off during GPT-5.2’s development, choosing to focus heavily on intelligence, reasoning, coding, and engineering tasks, while neglecting writing as a result of limited internal bandwidth.

According to Altman, that decision was intentional at the time. GPT-5.2 was designed to excel at professional and technical use cases, including complex reasoning, software development, and structured tasks such as spreadsheets and presentations. However, user expectations have continued to evolve, and many rely on ChatGPT not only for technical work but also for polished, client-facing writing. The feedback made it clear that the shift in priorities was noticeable in day-to-day use.

The contrast between GPT-4.5 and GPT-5.2 highlights how OpenAI’s focus changed over time. When GPT-4.5 launched in February 2025, OpenAI emphasised natural interaction and high-quality writing. The company described conversations with GPT-4.5 as feeling more natural and positioned the model as particularly useful for improving writing and refining tone.

GPT-5.2, by comparison, was introduced with a very different message. OpenAI framed it as a model family built for professional knowledge work, with strengths in coding, tool use, spreadsheet creation, presentation design, and managing complex multi-step projects. Writing was mentioned far less prominently, and technical writing was singled out as an area where GPT-5.2 Instant still needed improvement. Altman’s comments now suggest that the broader writing experience fell short of user expectations across the board.

ChatGPT has received numerous updates since GPT-5 launched in August 2025, including changes aimed at improving warmth and tone in instruction-following behaviour. OpenAI frequently adjusts model behaviour in response to feedback, and performance regressions in one area while improving another are not unusual. What stands out this time is how directly Altman acknowledged the trade-off and its consequences.

For users who rely on ChatGPT for drafting articles, client communications, or polished written output, the explanation offers some clarity. Model upgrades do not guarantee uniform improvements, and changes under the hood can affect workflows in unexpected ways. Treating major model updates like any other dependency change, by retesting prompts and maintaining fallbacks when output quality matters, may help reduce disruption.

Altman also reiterated his broader view of where OpenAI is headed. He said he believes the future lies in strong general-purpose models, adding that even systems optimised for coding and engineering should still be able to write well. However, he did not provide a specific timeline for when writing improvements will arrive. As with past updates, changes are likely to roll out gradually through point releases rather than appearing all at once.