OpenAI is ending the era of separate allowances for chatting and coding. Starting in December, ChatGPT Plus, Pro and Business plans will share a single token budget across ChatGPT and Codex, according to an updated OpenAI support page reported by The Decoder on November 1. The change arrives as OpenAI folds the agentic capabilities of ChatGPT Work into the main ChatGPT interface, a merger the company had already announced.
The mechanics matter more than the announcement. Usage will be measured in tokens, the units a model processes and generates, so more advanced models and more complex tasks will use more of the allowance. Chats with Instant models will not draw from the budget, which keeps quick, casual questions effectively flat-rate. Subscribers will be able to see their combined remaining usage and reset times in Settings, and when the pool runs dry the documented options are to wait for a reset, upgrade, or buy extra credits. OpenAI recently doubled usage limits for its higher-priced plans, apparently in preparation, but it has not published the size of the new shared budgets, an exact December start date, or the price of extra credits. An OpenAI engineer cautioned on October 28 that the merge was not rolling out yet and that the team would proceed carefully, so read the support page as a firm direction of travel, not a finished rollout.
For people who build with these tools, the practical consequence is immediate: a heavy afternoon of Codex agent work will soon compete with every ordinary ChatGPT question for the same allowance. Capacity planning becomes token budgeting. Teams will need to measure their actual burn by model and task, route routine work to Instant or cheaper models where quality allows, and reserve the shared pool for the advanced-model and agent runs that justify it. Anyone treating a subscription as all-you-can-prompt is budgeting on a model that ends in December, and with no published budget sizes, your own measured usage is the only reliable planning input.
The same week brought the mirror image of metering: marking. Anthropic told API customers by email that it is expanding its text watermark to four more models, Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6, with newer models such as Claude Opus 5.5 already carrying it and the Amazon Bedrock rollout possibly taking a few extra days. Anthropic describes the mark as an imperceptible pattern in word choice that changes neither meaning nor price, carries no information about the user, and cannot be switched off. The stated driver is the EU AI Act December deadline: models released after August 2, 2026 shipped with the watermark from day one, and older models are being retrofitted in stages, using a variant of Google DeepMind's SynthID text method, with a detection API in private preview.
There is, however, a documented loophole. On Opus 4.6 and Sonnet 4.6 the watermark applies only when the sampling parameters temperature, top_p and top_k are left at their defaults; changing any one of them is enough to produce unmarked output. Anthropic's own advice is blunt: if you rely on the watermark for your own compliance, keep those settings at their defaults. That caveat captures December's double lesson for builders. Usage is becoming one metered pool, and output is becoming provably labelled, but in both cases the details that decide whether you stay inside the lines are settings and numbers the providers control. Audit yours before December, not after.
