OpenAI says that starting in December, the usage allowance on Plus, Pro and Business plans will be shared across ChatGPT and Codex(opens in a new tab). One pool, metered in tokens, covers both the chat assistant and the coding agent; quick Instant chats sit outside it. OpenAI expects most users won't hit resets more often than today. It has not published how big each plan's pool will be, or the exact start date.
This reads like a consumer plan tweak. It is not. Business is the plan many companies buy for teams that sit outside a full enterprise contract, and Plus and Pro are what employees expense or buy themselves when the official tool falls short. From December, a seat on any of them stops being a flat allowance per product. An agent that refactors a codebase for an afternoon now spends the same tokens that a finance analyst needs for research chats. Heavy agent users will run dry first, and they are usually the people you most want using it.
It fits a wider shift in AI pricing: seats are turning into prepaid meters. A per-seat price still looks fixed on the invoice, but the work you get out of it now depends on which models people pick and how long their agents run. Two teams on the same plan can get very different amounts of work done, and the gap shows up as people waiting for resets or buying extra credits, not as a line item.
For the CIO and CFO. Before December, pull usage reports for every OpenAI seat you pay for, directly or through expenses. Split the heavy agent users onto usage-billed or higher tiers and leave light chat users where they are. Then set a per-team monthly ceiling, so an agent-heavy quarter shows up in your budget instead of as stalled work.
Action items
From December, ChatGPT and Codex will draw on one shared token allowance for Plus, Pro and Business seats. A seat stops being a flat allowance per product: an employee's coding agent now spends the same budget as their chat.
For the CIO and CFO
Audit usage on every OpenAI seat you pay for before December, and move agent-heavy users onto usage-billed tiers.
Researched and drafted by an automated workflow, then reviewed and edited by a human editor before publication. Every source is linked. See how we use AI here.
Also worth knowing
- OpenAI Developer CommunityUsage allowance will be shared across ChatGPT and Codex(opens in a new tab)
From December, Plus, Pro and Business seats draw one shared token pool for ChatGPT and Codex. Per-plan sizes are not yet published, so audit agent-heavy users before the switch.
- The RegisterA whopping 3% of boardroom execs spy productivity and costs savings from AI(opens in a new tab)
Gartner says only 3% of organizations have hit their AI productivity and cost targets, 38% are over budget on AI, and 85% still plan to spend more next year. Fund deployments that turn freed time into measured output.
- TechCrunchMistral releases weights for its trillion-parameter AI model, targeting cybersecurity(opens in a new tab)
The promised Mistral Large 4 weights are out, pitched for cyber defense on your own hardware. A credible self-hosted option for security teams that cannot send incident data to a vendor API.
- PYMNTSNew York Bill Would Raise AI Safety Regulations(opens in a new tab)
A new bill would add 72-hour risk disclosures, licensed independent audits and court-ordered limits on a model's deployment in the state to New York's RAISE Act. It targets large developers, but a model pulled from New York would strand the companies there that build on it.