AI usage limits, side by side
Every major AI tool limits you, and no two do it the same way. Here is how Claude, ChatGPT, Gemini, Cursor, Grok and Perplexity compare in 2026, and how to keep an eye on all of them at once.
Claude meters rolling 5-hour and weekly windows. ChatGPT caps each model separately. Gemini moved to compute quotas in 2026. Cursor bills in dollars. Grok limits per request kind, and Perplexity caps daily searches and Deep Research. None of them shows you much before you hit the wall.
The comparison table
| Provider | How it limits you | Window and reset | Native visibility | Deep dive |
|---|---|---|---|---|
| Claude | 5-hour rolling window plus a weekly cap; Opus has a separate weekly allowance on Max | Rolling 5h and 7d | Minimal until you are near the limit | Details |
| ChatGPT | Per-model message caps, plus weekly caps on the heaviest reasoning models | About 3h rolling, plus weekly | None until the wall appears | Details |
| Gemini | Compute-based quota since the May 2026 change | Rolling window plus weekly | A native usage page | Details |
| Cursor | Dollar and usage-based spend, with on-demand charges past the included amount | Monthly billing cycle | A dashboard, in dollars | Details |
| Grok | Quotas per request kind (Auto, Fast, Expert, Heavy, DeepSearch, media) | Rolling, per kind | Minimal | Details |
| Perplexity | Daily search allowance plus a separate Deep Research cap | Daily, plus weekly | Minimal | Details |
Visibility is the common thread: every provider tells you the least it can get away with. That is the gap tokenkarma fills, in one view across all of them.
How each provider limits you
Claude
Two rolling windows at once, a 5-hour and a weekly cap, with Opus on its own weekly allowance on Max. Claude usage limits in detail.
ChatGPT
Per-model message caps on a rolling window, with weekly caps on the heaviest reasoning models, and no counter until you hit one. ChatGPT usage limits in detail.
Gemini
Compute-based quotas since May 2026, opaque in compute units rather than messages. Gemini usage limits in detail.
Cursor
A dollar limit, not a message limit: an included monthly amount, then on-demand charges that are easy to miss. Cursor usage and spend in detail.
Grok and Perplexity
Grok meters per request kind, Perplexity caps daily searches and Deep Research separately. Grok and Perplexity limits in detail.
Frequently asked questions
Which AI has the most generous usage limits?
It depends on how you work, because the limits are not measured the same way. Claude uses rolling time windows, ChatGPT uses per-model message caps, Gemini uses compute quotas, and Cursor bills in dollars. The honest answer is to track your own usage on each, which is what tokenkarma is for.
Why do AI usage limits feel so unpredictable?
Most providers use rolling windows rather than fixed-clock resets, do not publish exact thresholds, and change them often. Heavy models, long context and attachments all drain quotas faster, so the same number of messages can hit a limit on one day and not the next.
Do ChatGPT and Claude share the same kind of limit?
No. Claude meters a shared budget across two rolling windows. ChatGPT meters each model separately with its own cap. That is why a ChatGPT and a Claude limit can feel completely different even at similar usage.
How can I see all my AI limits at once?
tokenkarma reads each provider’s live usage and shows them side by side, with reset times and an alert before you hit a wall. It runs as a browser extension and a Mac app. See the AI usage tracker.
Did AI usage limits change in 2026?
Yes, repeatedly. Anthropic adjusted Claude limits several times in 2026, Google moved Gemini to compute-based quotas in May 2026, and Perplexity reduced limits mid-May. Each provider page here notes a last-updated month.
Track every one of these in one place
tokenkarma reads your live usage across Claude, ChatGPT, Gemini, Cursor, Grok and Perplexity, with reset times and alerts before a wall.
Free to start. Works in your browser and on your Mac.
tokenkarma is independent and not affiliated with any AI provider. Limit mechanics reflect each provider’s published information as of June 2026 and can change.