Pricing

AI credits and usage limits, explained

Why limits exist at all

Every prompt costs the vendor real computing power. A flat $20/month only works if most subscribers use modestly — heavy users are subsidized by light ones. Limits are how vendors keep that math working. They're also why "unlimited" in AI marketing always has an asterisk: it means "a lot, within our fair-use definition," never literally infinite.

Claude: rolling windows

Claude meters usage on a rolling 5-hour window — your budget resets 5 hours after your first prompt in that window, not at midnight. There are also weekly caps. Max 5x ($100/month) gets roughly five times Pro's budget; Max 20x ($200/month) roughly twenty times. Anthropic doesn't publish exact message counts (they shift with model and message size), so the honest way to check your remaining budget is the /usage command, not a third-party chart.

One wrinkle: since mid-2026, heavy programmatic use — the Agent SDK, command-line sessions, GitHub Actions — can draw from a separate credit pool billed at API rates on top of your subscription. If you automate with Claude, read Anthropic's current terms.

ChatGPT: tiered multipliers

OpenAI is explicit that none of its plans offer unlimited usage. The tiers are multipliers on the Plus baseline: the $100/month Pro tier offers roughly 5x the usage of Plus (aimed at heavy Codex coding-agent use), and the $200/month tier roughly 20x. The free tier is genuinely limited to a handful of messages every few hours. If you're hitting Plus limits weekly, the question is whether the $100 tier's 5x headroom matches your actual overage.

Microsoft Copilot: it depends on your plan

Copilot metering varies by license type and is managed through the Microsoft 365 admin center for business plans. Individual Copilot Pro users get priority access during peak times rather than a published credit balance. Because Microsoft adjusts this quietly and per-plan, the reliable move is checking your admin center or account page rather than trusting a blog post's numbers — including this one.

How to stay under the caps

• Batch your work. Ten small prompts cost more overhead than two well-constructed ones. Think before you prompt.
• Use the right tier for the job. Burning your best model's budget on trivia is how you hit caps by Thursday. Save heavy usage for heavy work.
• Watch for the warning signs. Slower responses and "you're approaching your limit" notices mean the window is nearly spent — wrap up gracefully instead of starting something big.
• Don't pay up a tier on a hunch. Hit the cap three weeks in a row before upgrading. One heavy week is not a pattern.