Has GPT-5.4-Thinking become overly picky?

A growing number of users are reporting that OpenAI's GPT-4o model, particularly when using its 'Thinking' mode, has developed an annoyingly perfectionist streak. According to discussions on platforms like Reddit, the model has begun to frequently conclude its answers by nitpicking minor, often irrelevant details, even when such scrutiny isn't requested. More frustratingly, users note that the model contradicts its own conclusions across different chat sessions when given identical prompts. This behavior is reportedly occurring with the 'Efficient' personality setting enabled, without any custom instructions, suggesting a potential shift in the model's base tuning.

This perceived change is creating tangible workflow friction. Instead of providing a final answer and moving on, conversations get stuck in loops where the AI insists on highlighting minor caveats or potential edge cases. Users comparing the experience note that competitors like Anthropic's Claude 3.5 Sonnet more cleanly signal task completion, allowing for faster progression. The community speculation points to a possible recent update from OpenAI aimed at increasing precision or reducing hallucinations, which may have inadvertently made the model overly cautious and verbose in its reasoning process, prioritizing exhaustive analysis over practical utility.