OPENAI
When you prioritise code, writing pays the price
Sam Altman has said OpenAI fell short on writing quality in GPT-5.2.
Speaking at a developer town hall, Altman addressed feedback that GPT-5.2’s writing is harder to read than GPT-4.5.
He explained that OpenAI chose to focus development on reasoning, coding, and technical work, which meant writing received less attention.
Future GPT-5.x updates are expected to improve this.
The shift reflects how each model was positioned. GPT-4.5, released in early 2025, was promoted for natural interaction and writing support.
GPT-5.2, on the other hand, was presented as a stronger tool for professional knowledge work, with emphasis on coding, spreadsheets, presentations, and complex tasks.
Writing was included, but not as a core strength.
In brief:
GPT-5.2 focused on technical capability over writing
GPT-4.5 was stronger for natural language and drafting
Writing improvements are planned for future GPT-5.x releases
Smarter, but rougher around the edges
Altman’s comments help explain why some users have noticed a drop in writing quality.
OpenAI regularly adjusts its models, and gains in one area can come with trade-offs in another.
No timeline has been shared for writing improvements, though OpenAI typically rolls out changes gradually.
OpenAI is in a battle for the dumbest naming convention. Ditch the numbers, Sam. - MV