The 5-hour session limit
Your session starts with your first message and runs for five hours. Everything you send in that window counts against one bucket. The counter resets when the five hours are up — not five hours after your last message.
Independent guide · re-verified September 1, 2026
Anthropic’s help center gives approximate counts: around 45 messages per five hours on Pro, at least 225 on Max 5x and 900 on Max 20x. Here is what those numbers assume, what changed in 2026, and how to read your own meter.
The numbers
| Plan | Price | Messages per 5 hours | Models in chat | Weekly cap |
|---|---|---|---|---|
| Free | $0 | A few to ~15 short messages, variable | Sonnet 5, Haiku 4.5 | — |
| Pro | $20/mo ($17 annual) | ~45 | + Opus 5; Fable 5 via usage credits | Yes |
| Max 5x | $100/mo | ~225 | + Fable 5 (up to 50% of weekly limit) | Yes |
| Max 20x | $200/mo | ~900 | + Fable 5 (up to 50% of weekly limit) | Yes, separate Opus bucket |
The Pro, Max 5x and Max 20x counts come from Anthropic’s help-center articles on Pro and Max usage. They assume short messages on a light model and were written before the 2026 changes, so treat them as a floor, not a promise. Long chats, big files, Opus and Fable reduce them, and Settings → Usage is the only number that is definitely yours. Anthropic’s plan pages do not list models per plan; the models column reflects what the app showed at review time.
How it works
Your session starts with your first message and runs for five hours. Everything you send in that window counts against one bucket. The counter resets when the five hours are up — not five hours after your last message.
A second, longer cap sitting on top. Introduced in August 2025, it affects a small minority of subscribers, mostly heavy Claude Code users. On Max 20x, Opus is tracked in its own bucket separately from Sonnet.
Why this matters: the two have different fixes. Hit the session limit and a few hours clears it. Hit the weekly limit and waiting will not help — you need a lighter model, a smaller scope, or a higher tier. Settings → Usage shows both counters live, and that screen is the only number that is definitely yours.
Decoder
Four different walls look similar and need completely different responses.
| What Claude says | What it means | What to do |
|---|---|---|
| A reset time later today | Five-hour session cap | Wait until that time. It is accurate. |
| A reset date rather than a time | Weekly cap | Waiting hours will not help. Switch to Sonnet or Haiku, or wait for the stated date. |
| A limit named for a specific model | Opus-only cap | Other models still work. Switch and keep going. |
| “High demand” or capacity wording | Capacity, not your limit | Nothing of yours is consumed. Retry, or use a lighter model. |
| “I can’t help with that” | Policy, not usage | Waiting and upgrading change nothing. This is a content decision. |
Burn rate
Usage is measured in work, not messages. A short question and a 60-page PDF are not the same request.
Several times the cost of Sonnet per turn. This is the biggest single lever you have: draft on Sonnet, switch to Opus for the hard part.
The whole thread is reprocessed every turn, so message 50 costs far more than message 5. Start a new chat per task.
Each attachment is re-read from scratch. Put recurring material in a Project and reference it once.
One request fans out into many searches and reasoning steps behind the scenes.
Same pool as chat. A single agent run in the morning is the usual reason chat feels tight by lunch.
Changelog
Anthropic does not keep a public history of limit changes, so this log is maintained here.
| Date | Change |
|---|---|
| 28 Aug 2025 | Weekly limits introduced on top of the 5-hour window. Anthropic said it affected under 5% of subscribers. |
| 13–28 Mar 2026 | Temporary off-peak doubling of 5-hour limits, as reported by third-party trackers. Ended as scheduled. |
| 6 May 2026 | Claude Code five-hour limits permanently doubled; peak-hour throttling on Claude Code removed for Pro and Max (Anthropic announcement, 6 May 2026). |
| 1 Jul 2026 | Fable 5 launched on paid plans, included for up to 50% of weekly limits as a launch promotion (later extended to 19 Jul). |
| 20 Jul 2026 | Fable 5 included in Max, Team Premium and Enterprise Premium for up to 50% of the weekly limit. Pro and Team Standard use Fable 5 via usage credits, with a one-time credit granted. |
| Jul – 31 Aug 2026 | Temporary 50% boost to Claude Code weekly limits. First announced as ending 19 Aug, then extended to 31 Aug. |
| 14 Sep 2026 | Announced: the boost ends and weekly Claude Code limits settle at a permanent 25% above the pre-July baseline — about 17% below the boosted August level. Check Settings → Usage for your own meter. |
| 2026 | Usage credits: Pro and Max subscribers can prepay for extra usage, billed at API rates, that kicks in after the included allowance is used up. |
Alternatives
If a cap lands mid-task and you cannot wait for the reset, the practical fix is to have a second tool ready. None of these are affiliated with Anthropic; pick whichever fits the job.
Claude-style answers alongside ChatGPT, Gemini and Grok in one workspace, so one provider’s cap does not end your day.
Broad general assistant for writing, coding and everyday work.
Strongest inside Google Workspace and Google-context tasks.
Research answers with web sources and citations.
Disclosure: MultipleChat is built by the team that runs this site. It is listed as one option among several; nothing on this page is paid placement.
Go deeper
Sources
Source of the ~45 messages per 5 hours figure.
Source of the 225 and 900 message figures.
Current plan names and prices.
The only figure that is definitely yours.
Not affiliated with Anthropic. Limits and prices change; this page was last verified . Check Settings → Usage for your account.
FAQ
No. Anthropic's help center says around 45 messages every 5 hours, often more depending on capacity, plus a weekly cap on top.
Yes. Usage on claude.ai, the desktop app and Claude Code all draws from one shared pool, so a long agent run reduces what is left for chat.
Yes, on Pro and Max. Enable usage credits under Settings then Usage and add funds; they are billed at API rates and only kick in after your included usage is used up. Upgrading, waiting for the reset or switching to a lighter model remain the free options.
No. The Anthropic API is a separate product with its own rate limits, token pricing and billing in the console.
It breaks Anthropic's terms and puts your account at risk of review. Not worth it.
Only if you regularly exhaust Max 5x during real work. 20x costs four times as much as 5x and gives four times the messages, so there is no bulk discount at the top tier.