AI Usage Limits by Plan (September 2026): ChatGPT, Claude, Gemini, and Copilot on One Page

AI usage limits by plan: the short answer

As of September 2026, every major AI vendor meters paid plans on a rolling 5-hour window with a weekly cap on top, and only one of them publishes a real number. OpenAI prints estimates (10 to 100 GPT-5.6 Sol messages per 5 hours on Plus, 200 GPT-6 Pro messages a week on Pro $200). Anthropic prints multipliers (Pro is at least 5x Free, Max is 5x or 20x Pro). Google prints multipliers (AI Pro is 4x Free, Ultra is 5x or 20x Pro). Microsoft prints "usage limits apply" and a table of words like "extensive use." Every cell below comes from a vendor page I read on September 15, 2026, and where a vendor publishes nothing, the cell says "not published."

This page exists because the question "what are the usage limits" has four different answers depending on the logo, and each vendor only publishes its own. The detailed posts on ChatGPT usage limits, Codex usage limits, and Claude usage limits go deep on each meter. This is the table that puts them side by side.

Every plan, every vendor, one table

Last verified: September 15, 2026. This page is re-checked every Monday against the vendor pages listed in each section below.

How OpenAI meters ChatGPT

OpenAI is the only vendor that prints numbers, and it prints them for three separate meters. Everyday text chat is unlimited on every plan, subject to abuse guardrails. Reasoning with GPT-5.6 Sol has a limit OpenAI doesn't quantify on Plus, and when you hit it ChatGPT may continue on another model. GPT-6 Pro in Chat has hard weekly allowances (200 on Pro $200, 50 on Pro $100, 15 a month on a Business Standard seat, 50 a week on Premium). The full plan-by-plan version is on ChatGPT usage limits, and the seat math is on ChatGPT Business pricing.

How OpenAI meters Codex

ChatGPT Work and Codex share one agentic allowance, and OpenAI publishes it as estimated local messages per 5-hour window by model: 10 to 100 with GPT-5.6 Sol on Plus, 5 to 45 with GPT-6 Astra, 250 to 2,000 with Luna. Pro 5x and Business Premium multiply that by five, Pro 20x by twenty, and "weekly limits may also apply." Image generation burns the allowance 3 to 5x faster, cloud chats run Sol and can use more, and an active turn is allowed to finish when you hit the wall. Everything that quietly drains it is on Codex usage limits.

How Anthropic meters Claude

Anthropic publishes no message count for any plan and says so plainly: "there's no fixed message count." Every plan resets on a rolling 5-hour session window, paid plans add a weekly limit that resets at a fixed time assigned to your account, and claude.ai, Desktop, Claude Code, and Cowork all draw from one pool. What Anthropic does publish is relative: Pro is at least 5x Free per session, Max is 5x or 20x Pro, Team Standard seats are 1.25x Pro and Premium seats are 6.25x Pro. Paid plans can turn on usage credits at API rates, with bundles from $50 (billed $45) to $1,000 (billed $700). The mechanism, and why the missing number is the honest answer, is on Claude usage limits; the seat decision is on the Claude Team plan.

How Google meters Gemini

Since May 17, 2026, Gemini runs compute-based limits that "factor in the complexity of your prompt, the features you use, and the length of your chat," refreshing every 5 hours until you reach a weekly limit. Google publishes multipliers, not counts: AI Plus is 2x Free, AI Pro is 4x Free, AI Ultra is 5x or 20x Pro. Media generation, Deep Research, the Pro model, and Deep Think use more of the allowance, and subscribers who hit the wall can keep chatting on Flash-Lite or buy AI credits. Context windows are published: 32K tokens free, 128K on Plus, 1M on Pro and Ultra. The by-plan detail is on Gemini usage limits.

How Microsoft meters Copilot

Microsoft publishes the least. The Microsoft 365 Copilot Business add-on ($18 a user a month on the current promo, $21 list) carries no per-user prompt count on its pricing page; the metered part is agents, which draw Copilot Credits that admins cap per user per month, resetting on the first. Copilot Chat comes free with eligible Microsoft 365 plans and is "subject to service capacity." On the consumer side, Microsoft's own limits page lists Copilot chat as "extensive use" on Personal ($9.99), Family ($12.99), Premium ($19.99), and Pro ($99.99), with Voice and Vision graded "standard," "more," and "highest." No numbers anywhere. If your team lives in Word and Excel, that vagueness is a feature; if you want to budget usage, it's a problem.

Which plan to buy, by owner profile

Solo operator

Start at $20 and pick by the meter that will actually bind you. If your day is chat and documents, ChatGPT Plus gives you unlimited text and a Work allowance; Claude Pro gives you a 5-hour session that resets all day and includes Claude Code. If you're running an agent for hours, the $100 tier is where the real usage is: ChatGPT Pro $100 (5x) or Claude Max 5x. I'd skip $200 tiers until a $100 tier has locked you out twice in one week, and right now you can't buy ChatGPT Pro $200 anyway.

5-person team

Buy Standard seats for everyone and one Premium seat for whoever runs the agents. On ChatGPT Business that's 4 x $20 plus 1 x $100, or $180 a month billed annually, and the Premium seat has no 5-hour limit. On Claude Team it's the same $180 with the Premium seat at 6.25x Pro per session, and every seat's limit is per member, so one heavy user can't drain the others. Turn on workspace credits with a monthly cap before anyone hits a wall. Google AI Pro at $19.99 a head is cheaper, but it's a personal-account plan; for a business account you're looking at Workspace or Gemini Enterprise, where no limit is published.

20-person team

Seat-type mix is the whole decision. Run 30 days on Standard seats, read the usage analytics, then move the three or four people who hit limits to Premium. At 16 Standard plus 4 Premium, both ChatGPT Business and Claude Team land at $720 a month billed annually. If the team is already on Microsoft 365 Business Standard, the Copilot bundle at $23.50 a user is the cheapest all-in number on this page, but you're trading a published limit for "usage limits apply." The rule I'd set on day one, on any vendor: a monthly credit cap per seat type, so the first surprise is a notification and not an invoice.