Altman Admits GPT-5.2 Writing Flaws

Sam Altman has openly admitted that OpenAI made a mistake with the writing quality of GPT-5.2 after placing too much focus on technical abilities such as coding and reasoning.

He made the comments during a developer town hall session, where users raised concerns that GPT-5.2 produces text that feels harder to read and less natural compared with GPT-4.5.

Altman responded directly to this feedback, saying that the company “got it wrong” when it came to writing performance and acknowledged that the issue needs to be fixed in future updates.

According to him, the development team made a conscious decision to prioritise intelligence, logic, engineering and coding skills when building GPT-5.2. As a result, writing quality did not receive the same level of attention.

He explained that this was due to limited resources and time. When one area is strengthened, another can sometimes be overlooked, and in this case, writing suffered as a result.

Altman stressed that this change was not a permanent direction for the GPT model family. He believes future GPT-5 versions will restore and improve writing quality so that it performs better than GPT-4.5.

This difference becomes clearer when comparing how OpenAI introduced GPT-4.5 and GPT-5.2. GPT-4.5, launched in February 2025, was promoted as being more natural to talk to and especially useful for improving and refining written content.

At the time, OpenAI highlighted how GPT-4.5 could help users write more clearly and communicate ideas in a smoother and more human way.

GPT-5.2, however, was presented as a tool aimed mainly at professional and technical work. The launch focused on its ability to build spreadsheets, create presentations, write code and manage complex tasks that require multiple steps.

Although writing was mentioned in the GPT-5.2 announcement, it was mainly linked to technical documents rather than everyday or creative writing. This helps explain why users noticed a decline in readability.

For many people who use ChatGPT for reports, emails or client-facing content, the change in writing quality became quickly noticeable after the update.

This situation also shows how OpenAI regularly adjusts its models based on feedback. Since GPT-5 was first released, the company has already made several updates to tone, warmth and instruction-following.

What makes this case different is Altman’s honesty in admitting the trade-off. It is unusual for a tech leader to say so clearly that a feature became worse because of other development priorities.

For users who rely on ChatGPT for writing, this serves as a reminder that newer versions do not always improve every function at the same time. Model updates should be tested carefully before being used for important work.

Altman said he believes the long-term goal is to build strong general-purpose models that can handle many tasks well, including writing, reasoning and coding, rather than specialising too heavily in one area.

He did not give a specific timeline for when writing improvements will arrive in future GPT-5 releases. OpenAI usually introduces changes gradually through smaller updates rather than one major launch.

Overall, OpenAI appears aware of the issue and has signalled that improving writing quality will be a priority again. Future GPT-5 versions are expected to balance technical intelligence with clearer, more readable and more natural written output.