Claude Free vs Pro usage limits: one daily cap vs eight server buckets

Free is roughly 30 short messages on a daily-style reset (Anthropic publishes it as a range, around 30 to 100 in light usage, dropping to 20 to 30 the moment you attach files or run long conversations). Pro is roughly 45 short messages per rolling 5-hour window, plus a separate weekly compute budget. Anthropic's official phrasing is that Pro provides at least 5x the usage of Free per session, and that ratio holds across most days.

Not exactly. Anthropic does not publish a hard 24-hour reset for the Free tier, the way it publishes a sliding 5-hour window for Pro. In practice it behaves more like a daily-style budget that refills over a 4 to 8 hour band depending on your timezone and load. The point is Free is a single opaque limit; Pro is two clearly named limits (rolling 5-hour and weekly) that you can read off a JSON object.

Free is Claude Sonnet only (Sonnet 4.6 as of 2026), with no model picker. Pro adds the model selector and unlocks Opus 4.7 (heavier per byte, the model agentic loops actually want for hard refactors) and Haiku for cheap drafting. The model you pick on Pro materially affects which weekly sub-bucket fills first; Opus byte for byte fills seven_day_opus faster than Sonnet does seven_day_sonnet.

No. Claude Code is a Pro and Max feature. If you are on Free and want to run Claude as a local coding agent, you have to upgrade to Pro at $20 per month or use the API separately (the API is metered and not covered by either Pro or Max). The moment you turn on Claude Code is also the moment the rolling 5-hour bucket starts mattering, because agentic loops are the workload that pegs it before lunch.

Not in any useful way. claude.ai/settings/usage exists for Free accounts but the underlying /api/organizations/{org_uuid}/usage endpoint largely does not populate the bucket fields a Free user has. The Pro account version of that endpoint returns a JSON object with five_hour and seven_day populated; the Free version returns mostly empty. There is no rolling window to render because Free does not have one.

Eight optional fields on the same JSON object. ClaudeMeter's Rust deserializer at src/models.rs lines 18 to 28 names them: five_hour, seven_day, seven_day_sonnet, seven_day_opus, seven_day_oauth_apps, seven_day_omelette, seven_day_cowork, plus an extra_usage object for paid overage. On a typical Pro account, two of those populate (five_hour, seven_day). On a Max account with Claude Code, four to five populate. On Free, zero of them populate in any meaningful way. That structural difference, not the message-count multiplier, is what changes about your day.

By Anthropic's official phrasing, Pro is 'at least 5x' the usage of Free per session, plus a weekly compute budget that Free does not have at all. In practice it lands closer to 5x to 10x depending on what you do. A heavy claude.ai writer who only sends short prompts gets a clean 5x. A Claude Code user who triggers agentic loops with attachments and Opus selected gets much more value out of Pro because Free literally cannot run those workflows.

Because Pro replaces one opaque ceiling with two named ceilings (5-hour, weekly) plus paid overage. On Free you cannot really overspend; you hit the cap, you stop. On Pro you can: blow through the rolling 5-hour and not realize you have crossed onto metered overage until your card statement; or run an Opus-heavy week that pegs seven_day_opus on a Tuesday afternoon while seven_day overall is at 60 percent. ClaudeMeter exists for that gap. Free users do not need it (nothing to track); Pro users do (eight buckets, two undocumented).

Anthropic's March 2026 statement says the company is adjusting the rolling 5-hour window during peak hours of 5 to 11 a.m. Pacific on weekdays, while leaving overall weekly limits unchanged. That multiplier applies to Pro and Max five_hour buckets specifically, because those are the buckets that exist. Free users feel peak hour load mostly as 'priority access goes to paid users first,' not as a quota multiplier; the Free limit itself is not weighted hour by hour.

No. ccusage walks ~/.claude/projects/*.jsonl on disk and totals input_tokens + output_tokens on this machine. That is local Claude Code spend; it is a different question from server quota. ccusage cannot tell you whether you are on Free or Pro, cannot read either the 5-hour or the 7-day bucket, and cannot see browser-chat usage that is depleting the same buckets. Run ccusage for cost, run ClaudeMeter for limits. They answer different questions; many people on Pro run both.

If your Free workflow is occasional short prompts and you almost never run into the cap, no. If you find yourself rationing prompts, switching to ChatGPT mid-task because Claude is locked, attaching files and watching the daily cap halve, or wanting Claude Code at all, yes. Pro is $20 per month for at least 5x the per-session usage, the model picker, Projects, and Claude Code. The honest reason most people upgrade is Claude Code, not the message count.