Have you noticed anything strange lately with OpenAI’s models? I just stumbled upon a fascinating observation from a daily user who’s been tracking some significant changes in output over the past three days. Apparently, the 4o model is now consistently ‘thinking’ and taking multi-minute pauses before responding. But what’s even more intriguing is that the quality of the output has improved dramatically, especially for coding tasks.
For instance, the user was able to build a working cube game clone in just seven prompts, with 99% of the code completed on the first try. The SVG test results are also showing a much closer resemblance to the leaked GPT-5 results. It’s as if the model is now capable of producing higher-quality outputs, almost like it’s being routed to another model.
But what’s behind this mysterious shift? Is OpenAI testing a new model or experimenting with a different approach? The user suspects it might be an A/B test or a model router that’s allowing the usage of other models. Whatever the reason, the results are undeniably impressive.
I’d love to hear your thoughts on this. Are you experiencing similar improvements with OpenAI’s models?