Peak Hours and Usage Limits
Everything you need to know about AI rate limits, and how tokenkarma helps you navigate them.
What are peak hours?
AI providers serve from a finite pool of compute. When demand spikes, several of them officially reserve the right to tighten usage limits rather than degrade everyone's experience. That is what "peak hours" means in practice: the same plan, a smaller effective allowance, at the busiest times of day.
The mechanics are rarely a published timetable. More often the provider documents the principle and keeps the schedule and the multiplier to itself, which is exactly why heavy users get surprised mid-task.
Claude: the 2026 peak-hours timeline
Anthropic is the provider that has been most explicit about peak-hours behavior, and the policy moved twice in 2026:
- March 26, 2026: Anthropic technical staff member Thariq announced on X that session limits would tighten during weekday peak hours, defined as "5am–11am PT / 1pm–7pm GMT": "you'll move through your 5-hour session limits faster than before." The thread estimated about 7 percent of users would hit session limits they would not have hit before, and advised shifting token-intensive background jobs to off-peak hours. Notably, this window was announced on X and never published on an Anthropic support page.
- May 6, 2026: Anthropic officially announced it was "doubling Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans" and "removing the peak hours limit reduction on Claude Code for Pro and Max accounts", on the back of a compute deal with SpaceX.
Where that leaves Claude users today: the peak-hours reduction is officially gone for Claude Code on Pro and Max. For other surfaces, the only live official mention of peak hours is on Anthropic's Pro plan page, which describes Pro capacity as "at least five times the usage per session compared to our free service" specifically "during peak hours". No current official page publishes a time window, so treat any specific schedule you read elsewhere as historical.
What each provider officially says
Beyond Anthropic, most providers document demand-based limits in principle, not in schedule. Here is what is actually on their official pages as of June 2026:
| Provider | Official position on demand-based limits |
|---|---|
| Claude | Peak hours acknowledged on the Pro plan page; the Claude Code reduction was removed May 6, 2026 for Pro and Max |
| ChatGPT | Free-tier limits "can be dynamic and may vary based on factors that include market, system conditions, abuse-prevention guardrails, and individual usage" (official help center); no published schedule |
| Gemini | "Limits may change without notice, including due to capacity constraints", and some features "may be unavailable during periods of high demand" for users without a paid plan (official help center) |
| Cursor | No peak or demand-based language in the official docs: limits are monthly budget pools and spend caps |
| Perplexity | No official peak-hours documentation for consumer plans |
| Grok | "Limits can vary slightly by platform or subscription" (official FAQ); no demand-based policy published |
How to plan around peak hours
Whatever the current policy, the playbook for heavy users stays the same, and it is the same advice Anthropic's own staff gave when the reduction was live:
- Shift the heavy jobs off-peak. Long agent runs, batch processing and token-intensive background work do not care what time it is; your session windows do. Early afternoon or evening Pacific time has historically been the quieter side.
- Watch your resets, not the clock. What actually stops you is a window running out. Knowing the exact reset time for each limit turns "I hope this goes through" into a schedule.
- Keep a second model warm. If your main provider tightens at the hour you work, the cheapest insurance is knowing which of your other subscriptions has room right now.
- Re-check the policy monthly. Both 2026 Claude changes happened within six weeks of each other. Provider limit policy is now a moving target, which is half the reason this page exists.
How tokenkarma helps
tokenkarma tracks your usage windows in real time across Claude, ChatGPT, Gemini, Cursor, Grok and Perplexity, with the exact reset time for each limit and alerts before you hit a wall, whatever the provider's current peak policy is. When limits tighten with demand, you see your window filling faster in the numbers themselves, not in a surprise block message.
The AI Cost Strategist, coming next, will take your current usage and reset times into account when it recommends a workflow.
Updated June 2026. Every provider statement above links to its official source. tokenkarma is not affiliated with Anthropic or any AI provider.