OpenAI models may get more reasoning budget through the API than through Codex

OpenAI is apparently adjusting the reasoning capabilities of its models, with reports indicating that API users may access significantly higher performance than those utilizing Codex. Specifically, the upcoming GPT-6 Astra model reportedly boasts a reasoning capacity of 832 units via the API, compared to just 128 units through Codex. This discrepancy suggests that users relying on Codex may be operating with a constrained version of the model, possibly limiting their access to its full potential.

This development is significant for the AI/ML community as it raises questions about the availability and fairness of model performance across different platforms. Developers and researchers using Codex may inadvertently be at a disadvantage compared to those who invest in the API, potentially impacting the outcomes of their projects. If the measurements proving this disparity hold true, it could lead to increased scrutiny of OpenAI’s model deployment strategies and prompt discussions about equitable access to cutting-edge AI capabilities.