ChatGPT and Codex usage limits merging

Discussion overview

The thread treated the change less as a cleaner product integration than as a subscription-value cut hiding behind a friendlier ChatGPT-and-workflow announcement. Several paying users had built a deliberate split workflow: use ChatGPT for long reasoning, research, debugging logs, and reviewing diffs, then spend Codex capacity on implementation. Separate pools made that practical, especially for researchers feeding large simulation outputs into chat. A shared allowance threatens to turn an intensive coding session into a lockout from ordinary questions precisely when those questions are needed to inspect the work.

There was broad agreement that a merged limit is technically unsurprising: the same expensive underlying models are being consumed regardless of interface, and a unified budget can curb frivolous use of costly agentic or premium requests. One commenter also argued that normal chat uses so few tokens beside autonomous coding that the operational impact may prove negligible. But that observation sharpened the criticism rather than calming it: if casual chat is cheap, users asked why a high-priced subscription cannot preserve a small protected chat reserve. Readers read the move as evidence that generous AI subscription trials are ending and that OpenAI is closing an unprofitable gap. Many said the low switching cost makes rival hosted services or open models more attractive if shared caps materially reduce access.

Points of disagreement

The core fight is whether one shared quota is a sensible reflection of model costs or a stealth downgrade for premium subscribers. Defenders say chat and Codex ultimately draw on the same capacity, routine conversations consume little beside agentic coding, and merged limits discourage wasteful premium requests. Critics say decoupled access was a major reason to pay OpenAI, and exhausting coding capacity should not disable the chat, research, and debugging tools needed to finish the job.