**OpenAI chief executive Sam Altman said the company made a mistake in GPT-5.2 by allowing writing quality to decline as development efforts were concentrated on coding and reasoning, acknowledging user complaints and signaling improvements in future releases.**

![Altman Admits GPT-5.2 Writing Fell Short of Expectations](https://www.stanventures.com/news/wp-content/uploads/2026/01/Screenshot-2026-01-28-172918-300x242.png)

Speaking during a developer town hall on Monday evening, Altman addressed growing criticism that GPT-5.2 produces writing that is harder to read and less fluid than earlier versions, particularly GPT-4.5. Users had described the output as unwieldy and difficult to use for polished writing tasks.

Altman did not dispute the assessment. 

He said OpenAI “screwed up” the model’s writing quality and accepted responsibility for the regression, describing it as an unintended consequence of development priorities rather than a design goal.

His full remarks are available in the below video.

[https://www.youtube.com/live/Wpxv-8nG8ec?si=d_Jm0b1eV-fOGqg8](https://www.youtube.com/live/Wpxv-8nG8ec?si=d_Jm0b1eV-fOGqg8) 

## Development Focus Shifted to Technical Capabilities

According to Altman, OpenAI deliberately concentrated GPT-5.2 development on improving intelligence, reasoning, coding, and engineering tasks. 

He said limited development capacity meant that gains in those areas came at the expense of writing performance.

The decision reflects a broader push to make GPT-5.2 suitable for professional knowledge work, including software development, data handling, spreadsheet creation, presentations, and complex multi-step workflows.

Altman said the tradeoff was not intended to redefine OpenAI’s long-term direction and stressed that strong writing remains a core expectation for general-purpose AI systems.

## Contrast With GPT-4.5

When OpenAI introduced GPT-4.5 in February 2025, the company emphasized natural interaction and improved writing, describing the model as more conversational and better suited for drafting and editing text.

GPT-5.2, by contrast, was marketed as OpenAI’s most capable model family for advanced technical tasks. 

Writing received less prominence in launch materials, with improvements framed mainly around technical documentation rather than general prose.

Altman’s remarks suggest that users who relied on ChatGPT for client-facing writing or editorial work felt the impact of that shift more sharply than others.

## An Unusually Direct Acknowledgment

OpenAI regularly updates its models and adjusts behavior based on user feedback. What stood out in this case was the directness of Altman’s admission.

Public acknowledgments of regressions are uncommon, particularly when tied to deliberate resource allocation choices. 

Altman said improvements in one area can come at the cost of others during a given development cycle, underscoring that model upgrades do not guarantee uniform gains across all capabilities.

## Implications for Users

Users who rely on [AI writing tools](https://www.stanventures.com/blog/ai-content-writing-tools/) may see Altman’s explanation as clarification rather than reassurance.

His remarks suggest writing quality can vary as OpenAI balances development priorities. Those who need consistent tone or readability may need to adjust workflows or continue using earlier models where possible.

## What Comes Next

Altman said future GPT-5.x versions would improve writing and outperform GPT-4.5, while OpenAI continues to build out the model’s technical capabilities. He did not say when those changes would be released.

OpenAI generally rolls out changes through incremental updates rather than a single major revision.

The remarks reflect an acknowledgment that GPT-5.2 prioritized technical performance, with less emphasis placed on writing quality during development.

## What Users Should Keep in Mind

Here are a few considerations for users who rely on ChatGPT for drafting, editing, or other writing-focused tasks as OpenAI continues to iterate on its models.

1. Test writing-heavy workflows after any major model update rather than assuming output quality will improve.
2. Keep access to alternative models when clarity, tone, or polish matters for external communication.
3. Track point releases closely, as OpenAI often adjusts behavior through smaller updates.
4. Evaluate technical performance gains separately from language quality when assessing new versions.

## Key Takeaways

- Sam Altman openly admitted GPT-5.2 underperformed on writing quality.
- The regression resulted from prioritizing coding and reasoning capabilities.
- GPT-4.5 remained stronger for natural writing and tone.
- OpenAI plans to improve writing in future GPT-5.x releases.
- Model updates can introduce tradeoffs, not universal improvements.