ChatGPT and Codex Will Share One Usage Limit From December

Starting in December 2026, OpenAI is folding ChatGPT chat and Codex into a single shared usage allowance for Plus, Pro, and Business subscribers, measured in tokens instead of separate message and request counts. The change surfaced through a quiet October 31 update to OpenAI's usage-limits help page, not a big launch post, and spread on Hacker News and X before OpenAI said much about it publicly.

If you're someone who chats with ChatGPT during the day and leans on Codex for coding at night, this is the update to actually pay attention to. Here's what's confirmed, what's still a mystery, and how it stacks up against how Claude Code has quietly handled this exact problem for over a year.

What's actually changing in December?

There's no exact start date. OpenAI's help page says "starting in December" and doesn't name a day, and it doesn't say whether the rollout lands for everyone at once or rolls out in stages. Here's what is confirmed:

• Plus ($20/month), Pro ($200/month), and Business ($25/user/month) move to one combined usage allowance shared between ChatGPT and Codex
• Usage gets measured in tokens, not per-message or per-request counts
• Instant chat is excluded entirely and won't draw from the shared pool
• Settings will show one combined remaining balance plus the next reset time, instead of two separate meters

OpenAI's own wording on this: "most users will not have to wait for resets more often than they do now." Read that twice. It's an expectation, not a guarantee, and that's exactly the kind of sentence a company writes when it genuinely doesn't know yet how this plays out at scale.

Wait, didn't Codex already have its own limit?

It did, and this is the part a lot of the coverage skipped over. On Plus and Pro, Codex, ChatGPT Work's agentic features, and the Office add-ins (Word, Excel, PowerPoint) already draw from one shared "agentic" allowance. December's change doesn't invent sharing from nothing, it folds regular chat with advanced reasoning models into that existing pool.

So the real shift isn't "Codex and ChatGPT never shared before." It's that your everyday chat habit, the kind of thing most Plus subscribers do far more of than coding, now competes for the same budget that Codex sessions already compete for. That's a meaningfully bigger pool of activity getting squeezed through one meter.

Why would OpenAI do this?

My honest read: this is a capacity-management move dressed up as a UX simplification. "One number to watch in Settings" is a genuinely nicer experience on paper. But the practical effect for a lot of users is that heavy Codex usage now eats into the chat headroom they used to take for granted, and vice versa.

Coding agents burn tokens fast. A single Codex session that reads a few files, writes a patch, and runs tests can chew through far more tokens than an entire afternoon of back-and-forth chat. Pool those together under one cap, and the people most likely to get squeezed are exactly the developers OpenAI wants sticking around: the ones using both products hardest.

That's also probably why OpenAI is being so vague about numbers. Publishing "Plus gets X tokens a month" invites everyone to do the math on whether their current combined usage already blows past it. Easier to ship the merge quietly, watch what happens, and adjust the dials afterward.

What's still unknown

A fair bit, honestly:

• No published token amount for any plan
• No clarity on whether the existing 5-hour and weekly caps for Codex carry over unchanged or get rebuilt around the new token system
• No word on Free, Go, Team, Enterprise, or Edu plans
• No confirmation on whether the rollout is simultaneous or staggered by account

Developer reaction on X has mostly landed somewhere between "annoyed" and "wait and see." A few people pointed out the obvious workaround problem: stacking a second paid account to get extra Codex capacity doesn't help nearly as much anymore if chat usage from the same account now eats into that same pool.

How does this compare to Claude Code's usage limits?

Here's the thing that makes this OpenAI change feel less like innovation and more like catch-up. Anthropic has run a unified usage model for Claude Code since mid-2025: a rolling 5-hour session cap plus a separate weekly cap, with no split between "chat" and "agentic" allowances in the first place. We covered the mechanics and the recent 25% baseline increase here if you want the full picture.

Anthropic isn't perfectly transparent either, there's no hard published token ceiling for Pro or Max, just rough hour estimates that fluctuate with model and task. But at least the structure has been stable and predictable for over a year. OpenAI is mid-transition, which means Plus and Pro subscribers get to be the test case while the company figures out where the dials should sit.

Should you be worried about this?

If you're a light user, mostly chatting with occasional Codex use, probably not much changes for you. If you're someone who runs long agentic Codex sessions regularly and also treats ChatGPT as your daily thinking tool, pay attention to your usage once this ships. Settings will show the combined meter, and that's worth checking the first week after the rollout hits your account rather than assuming your habits carry over unchanged.

If the merged pool turns out to be too tight for how you actually work both products, the honest options are the same ones that exist today: scale back one side of your usage, upgrade to a plan with a bigger allowance, or buy extra credits when OpenAI opens that up for personal plans the way it already has for Business and Enterprise. For context on how OpenAI's agentic push got here in the first place, we wrote about ChatGPT Work when it launched, which is the other half of what's getting folded into this shared pool.

I'll say this plainly: I don't love usage-pooling changes that ship through a help-center edit instead of a clear blog post with real numbers. OpenAI has the data to tell Plus subscribers exactly how this will land for them. Choosing not to, at least for now, tells you the company isn't fully confident it won't sting.

Official confirmation on X

Developer Tibor Blaho was first to spot the hidden disclaimer string in the ChatGPT web app's checkout flow, days before OpenAI's help page update made the rounds more widely, a pattern that's become pretty familiar with how OpenAI quietly ships config changes ahead of any formal announcement.