I just stumbled upon a fascinating conversation on Reddit about ChatGPT’s GPT-5 model. Apparently, some users have noticed that the model’s reasoning time and steps have been significantly reduced. Specifically, when you ask GPT-5-main to ‘think harder,’ it used to take around 2-4 minutes to reason and produce around 50 reasoning steps. Now, it only takes about 1 minute and produces 15-20 reasoning steps.
The Reddit user who shared this discovery believes that the downgrade is a cost-saving measure. They claim that the router no longer routes to ‘gpt-5-thinking-high’ but instead only routes to ‘gpt-5-thinking-low.’ This raises some interesting questions about the trade-offs between AI model performance and cost. Are we willing to sacrifice some of the model’s capabilities to make it more affordable? And what does this mean for the future of AI development?
It’s worth noting that this is not an official announcement from the developers, and we don’t know for sure what’s behind this change. However, it’s an important conversation to have, especially as AI models become more integrated into our daily lives. What do you think? Are you concerned about the potential downgrading of AI models for cost savings?