ChatGPT and Codex Will Share One Quota Pool Starting in December

OpenAI quietly updated its help documentation without an official announcement: starting in December, Plus and Pro subscribers will see ChatGPT chat and Codex usage merged into a single shared quota. Every interaction—whether you are chatting in the web interface or writing code in Codex—will draw from the same token-based pool. The more advanced the model, the deeper the reasoning, and the more complex the task, the faster the quota drains.

The change effectively ends the era of nearly unlimited ChatGPT chat for paid subscribers.

What Is Changing

Until now, ChatGPT chat and Codex have used separate quota pools. For Pro subscribers, the chat pool was effectively unlimited: many users reported multi-hour conversations at maximum reasoning depth without ever hitting a cap. Codex, by contrast, was already tightly constrained.

Starting in December:

• Plus and Pro subscriptions draw from one combined token pool for both ChatGPT chat and Codex.
• Usage is measured in tokens, not messages or turns.
• More capable models and deeper reasoning consume tokens faster.
• The only exception is Instant Chat, the lowest-reasoning mode, which remains unlimited.

Instant Chat is the lightweight, fast-response mode that skips extended reasoning. Users who rely on high-reasoning chat for research, fact-checking, or long technical discussions will see those conversations counted against the same pool as their coding sessions.

Why Plus Users Are the Hardest Hit

Plus subscribers already had minimal Codex access. Their main practical benefit was near-unlimited high-quality chat. That perk disappears under the merged quota.

Community reaction was immediate. A Plus user filed a GitHub issue on the Codex repository asking OpenAI to keep chat and Codex quotas separate. Commenters noted that unlimited chat was the one tangible advantage Plus held over competing subscriptions. Others framed the timing as an unwelcome holiday surprise.

For heavy chat users—developers who use ChatGPT as a research partner, fact-checker, or long-form thinking tool—the change means every question now has to be weighed against the week's remaining token budget.

The 2023 History Lesson

When GPT-4 launched in March 2023, a $20 ChatGPT Plus subscription covered roughly 100 messages every four hours. The limit was cut to 50, then to 25 every three hours. Every prompt had to be carefully composed because a wasted request was a wasted allowance.

Over time, the limits loosened, then effectively disappeared as OpenAI competed for users. ChatGPT grew to roughly 1.2 billion weekly active users. The chat product became the growth engine, while agent and coding revenue remained smaller.

Now the math has flipped. Agent workflows, Codex, and long-horizon tasks consume far more compute per user than casual chat. OpenAI is approaching an IPO and needs revenue to justify a valuation that depends on both scale and margin. Unlimited chat at $20 per month, when a single high-reasoning session can cost more than the subscription in inference, is no longer sustainable.

Claude Was Already There

Anthropic launched its $20 Pro subscription in 2023 on a token-based model from day one, rather than a message-count limit. When Claude Code exploded in popularity in 2025, the token billing model transferred over without friction.

OpenAI, by contrast, still uses message-count limits for its most capable agent model in chat mode: $100 Pro tier allows 50 requests per week, and $200 allows 200. That pricing model is mismatched with agent workflows where a single task can run for hours. The December move brings ChatGPT closer to the token-based billing that Claude has always used.

This is not the first reversal. A month ago, OpenAI tried cutting the Pro 200 Codex quota from 20x to 10x, then reversed course after community backlash. The merged quota is a more structural change: it does not just rebalance Codex; it redefines what the chat product itself costs.

The Meter Effect

Economics literature has a concept called the meter effect: people prefer flat-rate plans over pay-per-use because watching the meter tick creates psychological friction. For years, ChatGPT chat felt like an unlimited data plan. Users could open the interface, ask anything, and never think about consumption.

After December, that mental model changes. Every high-reasoning prompt becomes a small cost decision. The comparison users are already making is to mobile data plans: once unlimited, now capped, with the same monthly price buying dramatically less.

What This Means for Developers

The practical implication is that heavy ChatGPT users need to plan around the token pool. High-reasoning chat sessions, long context threads, and Codex runs will compete for the same budget. The default assumption should be that the December quota is tighter than what Plus and Pro users experienced in November.

OpenAI has not published the exact token allocations for Plus and Pro under the merged system. The documentation update will likely carry the specific numbers when the change takes effect. For teams that build on ChatGPT chat for research or internal tooling, December is the month to measure current consumption and decide whether to shift some workloads to other models or increase the subscription tier.

November 30 marks ChatGPT's fourth birthday. Four years after a simple chat interface introduced millions of people to AI, the era of unlimited conversation is ending. Whether the pendulum swings back as inference costs fall is a separate question, but for now, the meter is going on.