# Claude usage limits: how the five-hour and weekly limits work, and how to get more

Every Claude plan's limits as of October 2026: five-hour and weekly resets, Fable and Opus caps, what each limit message means, why usage drains fast, and every way to get more.

Author: Ravi Parikh

Canonical: https://voyager.so/blog/claude-usage-limits

Published: 2026-10-20T10:45:00-07:00
Updated: 2026-10-20T10:45:00-07:00

Claude's Pro, Max and Team plans have two usage limits that run at the same time: a session limit that resets every five hours and a weekly limit that resets at a fixed time assigned to your account. Chat, Cowork, Claude Code and the desktop and mobile apps all draw on the same pool. Anthropic publishes no message or token counts; plans differ by multiples of Pro's five-hour allowance ([how usage and length limits work](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work), [Pro plan](https://support.claude.com/en/articles/8325606-what-is-the-pro-plan)). To see where you stand, open [Settings, then Usage](https://claude.ai/settings/usage), or run `/usage` in Claude Code.

The short version: the limit message names the limit and its reset time. A free reset, usage credits or an upgrade get you going at once; switching models helps only with a model-specific limit.

Plan facts come from Anthropic's help center, Claude Code documentation and pricing pages, and other vendors' pages, read signed out in the US on 20 October 2026. Where the post uses independent measurements or user reports, it says so. Anthropic changed these limits repeatedly in 2026, so treat the date as part of every number.

Jump to: [you hit a limit](#you-hit-a-limit-what-to-do-now), [check your usage](#how-to-see-where-you-stand), [the plans](#the-limits-on-each-plan), [getting more](#getting-more-claude), [using less](#why-usage-drains-fast-and-what-helps).

## You hit a limit: what to do now

Claude Code names the limit in its message. In chat, Settings, then Usage, shows which bar is full.

| What you see                                                       | What it means                                                               | What you can do now                                                                                                                 |
| ------------------------------------------------------------------ | --------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- |
| "You've hit your session limit · resets 3:45pm"                    | The five-hour allowance is used up, for every model                         | Wait, use a free reset, continue on usage credits, or upgrade                                                                       |
| "You've hit your weekly limit · resets Mon 12:00am"                | The weekly allowance is used up, for every model                            | Same; switching models does not help                                                                                                |
| "You've hit your Opus limit" or "You've hit your Sonnet limit"     | A cap on that model family only                                             | `/model` and pick a model outside that family                                                                                       |
| "Fable limit reached · continuing on Fable 5.1 uses usage credits" | Max's 50% Fable cap is reached                                              | Accept credits, or switch model                                                                                                     |
| "Fable 5.1 now uses usage credits"                                 | Your plan runs Fable on credits only (Pro, Team Standard)                   | Accept credits, or switch model; if it adds "the prompt to confirm went unanswered", answer the prompt where the session is running |
| "Usage limit reached · wrapping up"                                | The five-hour limit hit mid-response, and Claude Code is finishing the step | Nothing; it stops after the step, or continues on usage credits if they are on                                                      |
| "Usage credits required for 1M context"                            | An entitlement check for a `[1m]` model variant, not a used-up allowance    | `/model` and pick the variant without `[1m]`                                                                                        |
| "Server is temporarily limiting requests (not your usage limit)"   | A short server-side throttle                                                | Wait briefly; Claude Code retries on its own. If it persists, check [status.claude.com](https://status.claude.com)                  |
| "You've hit your ... spend limit"                                  | Usage credits reached a spend limit you or your admin set                   | Raise the limit, or wait for the plan reset shown                                                                                   |

Sources: [Claude Code error reference](https://code.claude.com/docs/en/errors#youve-hit-your-session-limit), [Claude Code costs](https://code.claude.com/docs/en/costs#when-a-developer-asks-about-a-limit), [wrap-up allowance terms](https://support.claude.com/en/articles/17040437-claude-code-wrap-up-allowance) and [usage and length limits](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work), which says an upgrade's limits apply right away. Read 20 October 2026.

**Automatic continue.** In an interactive Claude Code session signed in with a Claude plan, Claude Code waits in the open session and picks the task back up after the reset, from version 2.1.234. It is on by default; `Esc` cancels it, and it gives up after two repeated limit hits in a row. It does not start on its own when the reset is more than 24 hours away, which is common for the weekly limit; choose the wait from `/rate-limit-options` instead ([wait for a usage limit to reset](https://code.claude.com/docs/en/interactive-mode#wait-for-a-usage-limit-to-reset)). Leave the terminal open. The desktop app has its own "Auto-continue when limits reset" checkbox on the session-limit card, but not on the weekly one ([Claude Code errors](https://code.claude.com/docs/en/errors#youve-hit-your-session-limit)).

**A free reset.** Anthropic gives limit resets "occasionally to eligible plans". Each restores the five-hour or the weekly allowance at once, without moving your weekly reset day. Use one from Settings, then Usage, or from the "Reset for free" button on the limit message, on the web or in Claude Desktop; the button is not in Claude Code or the mobile apps ([what is a limit reset](https://support.claude.com/en/articles/17007452-what-is-a-limit-reset)). Pro, Max and Team users got one with Opus 5.5 on 22 September, which Anthropic said can be applied "any time until Oct 22" ([@ClaudeDevs](https://x.com/ClaudeDevs/status/2102438803013333469)). If you still have that reset, check the exact expiry day and time on the offer in [Settings, then Usage](https://claude.ai/settings/usage), and use it before then; after it expires, check for a new offer rather than counting on the September reset ([reset expiry rules](https://support.claude.com/en/articles/17007452-what-is-a-limit-reset)).

**The wrap-up allowance.** If the five-hour limit lands mid-response, Claude Code on Pro, Max or a Team Premium seat may keep working briefly to finish the step. Pro gets it once per weekly period; Max and Team Premium at each five-hour limit. It needs Claude Code 2.1.277 or later, is still rolling out, and counts toward your weekly limit ([wrap-up allowance](https://support.claude.com/en/articles/17040437-claude-code-wrap-up-allowance)).

To be told when a limit stops an unattended session, a `StopFailure` hook runs when a turn ends on an API error, and its matcher can filter for `rate_limit` ([hooks reference](https://code.claude.com/docs/en/hooks#stopfailure)). On a Mac, this in `~/.claude/settings.json` shows a notification:

```json
{
  "hooks": {
    "StopFailure": [
      {
        "matcher": "rate_limit",
        "hooks": [
          { "type": "command", "command": "osascript -e 'display notification \"Claude Code stopped at a rate limit\" with title \"Claude Code\"'" }
        ]
      }
    ]
  }
}
```

## How to see where you stand

**Settings, then Usage** is your account's meter, so it covers every surface and device. It shows the current session and the time left, the weekly limit and its reset time, a separate Fable bar where your plan includes Fable, usage credits spent this month, and any free reset on offer ([usage limit best practices](https://support.claude.com/en/articles/9797557-usage-limit-best-practices)).

**`/usage` in Claude Code** shows the same plan bars plus where the usage went: the share that went to each skill, subagent, plugin and MCP server, flags for behaviours such as long context or cache misses when one accounts for 10% or more of recent usage, and the heaviest `/loop` tasks. Press `d` or `w` for the last 24 hours or 7 days. That breakdown is "computed from local session history on this machine, so usage from other devices or claude.ai is not included" ([Claude Code costs](https://code.claude.com/docs/en/costs#plan-usage-breakdown)). The `Total cost` line is a list-price estimate meant for API users and, on a plan, "isn't relevant for billing purposes".

**`/status`** shows which account you are signed in with. Check it first if usage looks wrong: Claude Code can use an `ANTHROPIC_API_KEY` from your environment and bill the API instead of your plan ([authentication](https://code.claude.com/docs/en/authentication)).

**The status line** can keep both meters on screen. Since version 2.1.80, Claude Code passes `rate_limits.five_hour` and `rate_limits.seven_day` to status line scripts for Pro and Max subscribers, each with a `used_percentage` and a `resets_at` time ([status line docs](https://code.claude.com/docs/en/statusline#rate-limit-usage), [v2.1.80](https://github.com/anthropics/claude-code/releases/tag/v2.1.80)). The quickest setup is to ask for it:

```text
/statusline show my 5-hour and 7-day usage limit percentages and when each resets
```

Claude Code writes the script to `~/.claude/` and updates your settings; the docs also include a ready-made script ([status line docs](https://code.claude.com/docs/en/statusline#use-the-%2Fstatusline-command)). In the desktop app, the usage ring next to the model picker shows the same thing.

**ccusage**, an open-source tool, answers a different question: what your local logs would have cost at API list prices. `npx ccusage@latest claude daily` reports Claude Code per day and `npx ccusage@latest blocks` per five-hour block ([ccusage](https://github.com/ccusage/ccusage)). It sees only the machine you run it on, and by default Claude Code's CLI removes transcripts older than 30 days unless you raise `cleanupPeriodDays` ([settings reference](https://code.claude.com/docs/en/settings-reference#cleanupperioddays)).

## The limits on each plan

| Plan                       | Price (US, October 2026)                        | Five-hour allowance          | Weekly limit    | Fable 5 and 5.1               |
| -------------------------- | ----------------------------------------------- | ---------------------------- | --------------- | ----------------------------- |
| Free                       | $0                                              | Baseline, varies with demand | Not documented  | Not available                 |
| Pro                        | $20 a month, or $200 a year                     | At least 5x Free             | Yes             | Usage credits only            |
| Max 5x                     | $100 a month                                    | 5x Pro                       | Yes             | Up to 50% of the weekly limit |
| Max 20x                    | $200 a month                                    | 20x Pro                      | Yes             | Up to 50% of the weekly limit |
| Team Standard seat         | $25 a month, or $20 billed annually             | 1.25x Pro                    | Yes, per member | Usage credits only            |
| Team Premium seat          | $125 a month, or $100 billed annually           | 6.25x Pro                    | Yes, per member | Up to 50% of the weekly limit |
| Enterprise (current plans) | $20 a seat a month, billed annually, plus usage | None included                | None included   | Billed at API rates           |

Sources: [Claude pricing](https://claude.com/pricing) for prices, "at least 5x more usage per 5-hour session than Free" and model access; [Free plan](https://support.claude.com/en/articles/8114491-get-started-with-claude), [Pro](https://support.claude.com/en/articles/8325606-what-is-the-pro-plan), [Max](https://support.claude.com/en/articles/11049741-what-is-the-max-plan) and [Team](https://support.claude.com/en/articles/9266767-what-is-the-team-plan) for allowances; [Fable on your plan](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan); [Enterprise](https://support.claude.com/en/articles/9797531-what-is-the-enterprise-plan), whose current seats have "no per-seat usage limits and no included token allowance". Read 20 October 2026. Team needs at least two seats and self-serve Enterprise 20; older seat-based Enterprise contracts keep Team-style allowances until renewal. Free includes Sonnet and Haiku, not Opus or Claude Code.

**The multipliers describe the five-hour session, not the week.** Max 20x is "20 times the Pro plan's per-session usage allowance" ([Max plan](https://support.claude.com/en/articles/11049741-what-is-the-max-plan)). Anthropic publishes no weekly ratio, and independent estimates suggest the weeks scale less (see [how much a plan holds in a month](#how-much-a-plan-holds-in-a-month)).

**One pool for every surface.** "\[Y]our usage of all different Claude product surfaces (claude.ai, Claude Code, Claude Desktop) counts towards the same usage limit" ([usage and length limits](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work)). That includes Claude Code in IDEs ([Claude Code with Pro or Max](https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-or-max-plan)) and Claude in Chrome ([Chrome troubleshooting](https://support.claude.com/en/articles/12902405-claude-in-chrome-troubleshooting)), and other apps that run Claude Code on your Claude sign-in, such as Voyager, the desktop app our team builds; those runs count against your plan as they would in a terminal.

**The Fable cap is a ceiling inside the weekly limit, not extra room.** On Max, "you can never use more than your weekly limit", and Fable models use the limit "faster than other Claude models" ([Fable on your plan](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan)). On Pro and Team Standard, Fable runs on usage credits from the first request, a change that took effect on 20 July 2026.

**Usage limits are not length limits.** A length limit is the context window of one conversation: up to 1M tokens on the newest models, 500K or 200K on others ([context window on paid plans](https://support.claude.com/en/articles/8606394-how-large-is-the-context-window-on-paid-claude-plans)). A new chat fixes a length limit; it does nothing for a usage limit.

**What resets when.** The five-hour session resets "on a rolling five-hour session window" ([pricing FAQ](https://claude.com/pricing)); Anthropic does not document exactly when a window starts, so go by the reset time it shows. The weekly limit resets "at a fixed time each week that is assigned to your account", unrelated to when you subscribed or started using Claude ([Pro plan](https://support.claude.com/en/articles/8325606-what-is-the-pro-plan)). The Fable cap resets with the weekly limit, and the Opus and Sonnet limits at the time in their message ([Claude Code errors](https://code.claude.com/docs/en/errors#youve-hit-your-session-limit)). Usage counts against both meters at once, so "\[a] single burst of heavy activity, such as a large workflow fanout, can exhaust the weekly allowance before the session window resets".

## Getting more Claude

There are three different things you can buy: more of your own plan, the same models on a separate pay-as-you-go meter, and another company's product that includes Claude models under its own rules ([covered further down](#claude-in-other-companies-plans)). Only the first keeps your Claude Code workflow exactly as it is; the second keeps Claude Code but changes how you sign in; the third mostly means working in another tool.

| Your situation                                    | Usually the best first move                                                                                                            |
| ------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------- |
| You hit the five-hour limit now and then          | Wait with automatic continue, or use a free reset; turn on usage credits with a low monthly spend limit for the days it matters        |
| You hit the five-hour limit on most busy days     | Move up a plan: Max 5x is five times Pro's session, Max 20x four times Max 5x's                                                        |
| You hit the weekly limit most weeks               | Move up a plan, knowing weekly room grows less than the name suggests: about twice from Max 5x to Max 20x for one Opus coding workload |
| Max 20x still runs out                            | Usage credits, bought as bundles to save up to 30%                                                                                     |
| Scripts, cron jobs or `claude -p` eat your limits | Run them on an API key; on Max or Team, the included monthly API credits pay for this                                                  |

The weekly ratio is from [SemiAnalysis's estimates](https://newsletter.semianalysis.com/p/anthropic-subscriptions-offer-5x) below; the rest is from the sources in this section.

Whether to upgrade or pay for overflow is arithmetic. On monthly billing, Pro to Max 5x costs $80 a month more and Max 5x to Max 20x $100 more ([pricing](https://claude.com/pricing)). Compare that step with what you actually pay for credits in a month, after any bundle discount: below the step, overflow is cheaper; above it, upgrade, provided the bigger plan would absorb the work. Someone on Pro paying $10 a month for credits spends $30 in total, against $100 for Max 5x.

For a sense of scale, consider 30 Sonnet 5.5 requests starting with 100,000 tokens already cached. Each request adds 10,000 new user or tool-result tokens and generates 2,000 output tokens, all retained in the conversation. Assume a five-minute cache, with every request starting less than five minutes after the previous one started, no compaction or cache invalidation, and all new input written to cache. This matches the default cache lifetime on usage credits; the lifetime runs from the request's start, including time spent generating its response ([Claude Code cache defaults](https://code.claude.com/docs/en/prompt-caching#which-ttl-each-request-gets), [API cache lifetime](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#how-prompt-caching-works)).

The first request reads 100,000 cached tokens and writes 10,000. Each later request also writes the preceding 2,000-token reply, so it writes 12,000 tokens; cached reads grow from 110,000 on request two to 446,000 on request 30 ([how conversation caching works](https://code.claude.com/docs/en/prompt-caching#how-the-cache-is-organized)). At standard global [API list prices](https://platform.claude.com/docs/en/about-claude/pricing), read 20 October 2026:

| Component                | Tokens across 30 requests                          | Cost calculation                                |
| ------------------------ | -------------------------------------------------- | ----------------------------------------------- |
| Cache reads              | 100,000 + 29 × (110,000 + 446,000) / 2 = 8,162,000 | 8.162 million × $0.10 = $0.8162                 |
| Five-minute cache writes | 10,000 + 29 × 12,000 = 358,000                     | 0.358 million × $2.50 = $0.895                  |
| Output                   | 30 × 2,000 = 60,000                                | 0.06 million × $10 = $0.60                      |
| Total                    |                                                    | $0.8162 + $0.895 + $0.60 = $2.3112, about $2.31 |

This excludes the earlier cost of warming the initial 100,000-token cache and assumes no separate tool fees, discounts or pricing modifiers. The last reply is charged as output but never sent back as input in these 30 requests.

### More of your own plan

* **Usage credits.** On Pro and Max, turn them on in Settings, then Usage, add funds (or turn on auto-reload), and set a monthly spend limit. On Team, an owner enables and funds them in Organization settings, then Usage, and members can then use them ([Team usage credits](https://support.claude.com/en/articles/12005970-manage-usage-credits-for-team-and-seat-based-enterprise-plans)). Then at any limit "\[y]our requests switch to pay-as-you-go pricing on usage credits and keep working", billed at standard API rates, until you reach a monthly spend limit you set or your prepaid balance runs out. In most places they do not expire; in some jurisdictions, such as Japan, they expire six months after purchase ([usage credits](https://support.claude.com/en/articles/12429409-manage-usage-credits-for-paid-claude-plans)). In Claude Code, `/usage-credits` opens the same settings ([Claude Code costs](https://code.claude.com/docs/en/costs#add-usage-credits-to-your-subscription)).
* **Bundles.** $45 buys $50 of credits, $200 buys $250 and $700 buys $1,000. Pro and Max subscribers can buy up to $2,000 of discounted credit value a month and Team owners $3,000; beyond that, credits cost the standard rate ([usage bundles](https://support.claude.com/en/articles/14246112-buy-usage-bundles)).
* **Upgrading.** The new plan's limits "apply right away" ([usage and length limits](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work)), and upgrades are prorated ([Max plan](https://support.claude.com/en/articles/11049741-what-is-the-max-plan)). Max 20x is the largest individual plan.

One catch on credits: once you are past your limit, Claude Code shortens the main conversation's cache from one hour to five minutes by default, so every pause over five minutes re-processes the conversation at your expense (see [the cache lifetime](#why-usage-drains-fast-and-what-helps)).

### The same models on a separate meter

* **An API key.** A Claude Console account bills per token, separately from any plan; purchased API credits expire after a year ([paying for API usage](https://support.claude.com/en/articles/8977456-how-do-i-pay-for-my-claude-api-usage)). Claude Code uses the key when you sign in with it or approve an `ANTHROPIC_API_KEY` ([authentication](https://code.claude.com/docs/en/authentication)). The API has its own limits: the standard Start, Build and Scale tiers cap monthly spend at $500, $1,000 and $200,000, new organizations may begin on a lower Evaluation tier, and every tier has per-minute token limits ([rate limits](https://platform.claude.com/docs/en/api/rate-limits)). Batch requests are half price if you can wait ([API pricing](https://platform.claude.com/docs/en/about-claude/pricing)).
* **Max and Team API credits.** Since 7 October, Max 5x includes $100 a month of API credits and Max 20x $200; Team gets $20 per Standard seat and $100 per Premium seat, pooled and capped at $500. After seven days on the plan, a Max subscriber claims them in Settings, then Billing, and a Team owner in Organization settings, then Billing, by linking one Claude Console organization. They cover the Claude API, Managed Agents, the Agent SDK, and `claude -p` you run yourself with an API key from that organization. They do not cover interactive Claude Code, usage credits, or Claude on Bedrock, Vertex AI or Foundry; unused credits expire each billing cycle; and they "don't change your usage limits" ([monthly API credits](https://support.claude.com/en/articles/17154008-monthly-api-credits-for-max-and-team-plans)).
* **Cloud providers.** Amazon Bedrock, Google Cloud and Microsoft Foundry sell the current Claude models on your cloud bill, and Claude Code runs against each ([Bedrock](https://code.claude.com/docs/en/amazon-bedrock), [Google Cloud](https://code.claude.com/docs/en/google-vertex-ai), [Foundry](https://code.claude.com/docs/en/microsoft-foundry)). This suits companies with cloud commitments. Pin the model, because Claude Code's default there "can lag the newest release" ([Bedrock](https://code.claude.com/docs/en/amazon-bedrock)).

One thing to avoid: using a second Claude account to get around your limits. Anthropic's terms bar sharing an account, and its wrap-up allowance terms name "an attempt to get around usage limits (for example, through multiple accounts or automation designed to exploit the allowance)" as grounds to withhold that allowance ([Consumer Terms](https://www.anthropic.com/legal/consumer-terms), [wrap-up allowance](https://support.claude.com/en/articles/17040437-claude-code-wrap-up-allowance)).

## Why usage drains fast, and what helps

Most surprises come from one fact: every turn re-sends the whole conversation. In Claude Code, "each time Claude uses tools it sends another request carrying that batch of tool results", so "a one-line question in a session that has been open all day still draws usage for the whole conversation" ([Claude Code costs](https://code.claude.com/docs/en/costs#why-usage-climbs-in-a-long-session)). Prompt caching makes the re-read cheap, but only while the cache is warm.

**Start from what `/usage` tells you.** Its breakdown flags any behaviour behind 10% or more of recent usage, and since version 2.1.299 it also warns about max effort, workflows, always-on context and large tool results ([Claude Code costs](https://code.claude.com/docs/en/costs#plan-usage-breakdown), [v2.1.299](https://github.com/anthropics/claude-code/releases/tag/v2.1.299)). Read it like this:

| What `/usage` shows                                                          | Likely cause                                           | First thing to try                                                                                 |
| ---------------------------------------------------------------------------- | ------------------------------------------------------ | -------------------------------------------------------------------------------------------------- |
| Cache misses flagged, or `Prompt cache (main)` shows misses and a cold cache | Gaps longer than the cache lifetime, or model switches | Fewer long pauses on big sessions; compact before a break; on credits, a one-hour TTL              |
| Long context flagged                                                         | One conversation carrying hours of history             | `/clear` between tasks; `/compact` at natural breaks                                               |
| A large subagent, workflow or plugin share                                   | Fan-out work, each agent with its own context          | Smaller model for subagents; if ultracode is on, `/effort ultracode off` stops automatic workflows |
| A `/loop` row with many runs                                                 | A scheduled task firing while you are away             | Stop it, or lengthen its interval                                                                  |
| One MCP server with a large share                                            | Big tool results from that server                      | Ask for narrower results, or use a CLI tool instead                                                |
| Max effort or large tool results flagged                                     | Effort or output size, not context                     | Lower the effort; trim logs before Claude reads them                                               |

**The cache lifetime is the hidden variable.** By default, on a Claude plan within your included usage, Claude Code asks for a one-hour cache on the main conversation, while subagents, workflows and compaction get five minutes. On usage credits, as on an API key, the main conversation also drops to five minutes ([prompt caching in Claude Code](https://code.claude.com/docs/en/prompt-caching#which-ttl-each-request-gets)). The first message after a longer gap re-processes the whole context.

At API prices, which is what usage credits bill, here is the input cost of the next message in a 300,000-token Opus 5.5 conversation, as of 20 October 2026 ([API pricing](https://platform.claude.com/docs/en/about-claude/pricing)):

| Your next message                                              | Input cost |
| -------------------------------------------------------------- | ---------- |
| Within the cache lifetime (cache read, $0.20 per million)      | $0.06      |
| After a five-minute cache expired (re-write at $5 per million) | $1.50      |
| After a one-hour cache expired (re-write at $8 per million)    | $2.40      |

Our arithmetic: one cold resume costs as much as 25 to 40 warm re-reads. Anthropic does not publish how cache reads and writes weigh on the subscription meters, so the ratio inside your plan may differ, but Claude Code's own behaviour shows the cost is real. Version 2.1.304, released 19 October, warns on Pro and Max before resuming an inactive conversation that "would use a noticeable share of your 5-hour usage limit", and 2.1.299 compacts terminal conversations over 200K tokens that sit idle for about 54 minutes, before the one-hour cache expires ([v2.1.304](https://github.com/anthropics/claude-code/releases/tag/v2.1.304), [v2.1.299](https://github.com/anthropics/claude-code/releases/tag/v2.1.299)).

Since 13 October, users have filed GitHub issues reporting that interactive Claude Code sessions on a subscription were writing five-minute rather than one-hour cache entries, with reports on versions through 2.1.304 on 20 October ([#101685](https://github.com/anthropics/claude-code/issues/101685), [#103673](https://github.com/anthropics/claude-code/issues/103673)); other commenters on the first report saw cache hits well after five minutes, so the label alone may not mean early expiry. A separate report ties a Max 20x account's use of 54% of its weekly limit in under a day to subagent caches expiring during long commands ([#103309](https://github.com/anthropics/claude-code/issues/103309)). All three reports remained open, with no Anthropic response posted, when we rechecked them on 20 October. If your usage jumped this month, the `Prompt cache (main)` line in `/usage` shows the reported cache lifetime and recent misses ([Claude Code costs](https://code.claude.com/docs/en/costs#prompt-cache-statistics)).

**What helps**, mechanism by mechanism:

1. **`/clear` between unrelated tasks.** Every later turn stops paying for the old task, and clearing costs nothing. Use `/rename` first if you may want to `/resume` it ([Claude Code costs](https://code.claude.com/docs/en/costs#manage-context-proactively)). In chat, start a new conversation.
2. **Compact while the cache is warm, not after a break.** A mid-session `/compact` reads the conversation from cache and "costs a fraction of what the context size suggests"; after a break longer than the cache lifetime, it re-processes the full history ([compacting the conversation](https://code.claude.com/docs/en/prompt-caching#compacting-the-conversation)). Before a long break on unfinished work, compact or have Claude write a short handoff note; if the work is done, `/clear`. To abandon a wrong turn, `/rewind` goes back to a prefix that is already cached.
3. **Choose the cache lifetime on credits.** If you work on usage credits or an API key and often pause between five minutes and an hour on a large conversation, setting `promptCacheTtl` to `1h` in your settings usually costs less: the higher write rate applies only to each turn's new tokens, while a miss re-writes the whole conversation. On Opus 5.5, writing 10,000 new tokens a turn at the one-hour rate costs $0.03 more than at the five-minute rate; one miss on a 300,000-token conversation costs $1.50. Past an hour, both expire. `subagentPromptCacheTtl` does the same for subagents; both need Claude Code 2.1.242 or later ([choose the TTL yourself](https://code.claude.com/docs/en/prompt-caching#choose-the-ttl-yourself)).
4. **Leave a cheaper model on for routine work.** Anthropic does not publish how each model weighs on your limits, but when it priced Opus 5.5 below Opus 5 it said "your 5-hour and weekly limits go 25% further" ([@ClaudeDevs](https://x.com/ClaudeDevs/status/2102438803013333469)). At API prices, Sonnet 5.5 is half of Opus 5.5 ($2 and $10 per million input and output tokens against $4 and $20) and Fable 5.1 two and a half times ($10 and $50) ([API pricing](https://platform.claude.com/docs/en/about-claude/pricing)). Judge it by the total for an acceptable result, correction turns included. `/model opusplan` plans with Opus and executes with Sonnet, but each switch between plan and execution starts a fresh cache ([switching models](https://code.claude.com/docs/en/prompt-caching#switching-models)). On Max, keep Fable for work that needs it.
5. **Lower the effort.** "Low and Medium work well for routine tasks and stretch your usage further" ([model, effort and thinking settings](https://support.claude.com/en/articles/8664678-change-the-model-effort-and-thinking-settings)). Thinking cannot be turned off on the 5.5 models or Fable, so effort is the dial ([Claude Code costs](https://code.claude.com/docs/en/costs#adjust-extended-thinking)).
6. **Watch subagents, workflows and teams.** Every subagent sends its own requests, and with ultracode on, Claude plans a workflow for every substantive task; turn it off with `/effort ultracode off` ([workflows](https://code.claude.com/docs/en/workflows#let-claude-decide-with-ultracode)). Set `model: haiku` for simple ones; agent teams "use approximately 7x more tokens than standard sessions" when teammates run in plan mode, so shut teammates down when they finish ([Claude Code costs](https://code.claude.com/docs/en/costs#manage-agent-team-costs)). Session-scoped `/loop` tasks, messages from your other sessions and goal check-ins also send your full context while you are away ([why usage climbs](https://code.claude.com/docs/en/costs#why-usage-climbs-in-a-long-session)).
7. **Trim what loads every turn, and scope the task.** Keep `CLAUDE.md` under about 200 lines and move workflow instructions into skills, which load only when used. MCP tool definitions are deferred by default, so check `/context` before blaming a server; CLI tools such as `gh` are still leaner ([reduce token usage](https://code.claude.com/docs/en/costs#reduce-mcp-server-overhead)). Name the file and function rather than asking Claude to "improve this codebase", and use plan mode before changes that touch several files ([work efficiently](https://code.claude.com/docs/en/costs#work-efficiently-on-complex-tasks)).
8. **In chat, use projects and fewer tools.** "\[C]ontent in projects is cached and counts less against your limits when reused" ([usage limit best practices](https://support.claude.com/en/articles/9797557-usage-limit-best-practices)), and "Tools and connectors are token-intensive", so turn off connectors a conversation does not need. Long chats that Claude summarizes to continue "consume more of your usage limit" ([usage and length limits](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work)).

Two more drains to know: switching models mid-conversation re-reads everything without cache, and fast mode in Claude Code is billed to usage credits even when your plan has usage left ([fast mode](https://code.claude.com/docs/en/fast-mode)). Then check the flags in `/usage` to see whether the change worked.

## Claude in other companies' plans

These do not add to your Claude plan; they are other tools with their own Claude access. Most of the ones below charge for Claude at or near Anthropic's API list prices, which makes them a convenience rather than a discount; the exceptions are a fixed quota or a bonus allowance.

| Product                                      | Price (US, October 2026)                        | Claude models                                              | How Claude usage is metered                                                                 |
| -------------------------------------------- | ----------------------------------------------- | ---------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
| Google Antigravity on Google AI Pro or Ultra | $19.99 (Pro); $99.99 or $199.99 (Ultra) a month | Sonnet 5.5 and Opus 5.5 on paid Pro (not trials) and Ultra | A separate five-hour and weekly quota for Claude and GPT models; no token amounts published |
| GitHub Copilot Pro+ or Max                   | $39 or $100 a month                             | Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1                 | AI credits at Anthropic's list prices; $70 a month on Pro+, $200 on Max                     |
| GitHub Copilot Pro                           | $10 a month                                     | Sonnet 5.5, Haiku 5.5                                      | $15 of AI credits a month                                                                   |
| Kiro (Amazon)                                | $20 to $200 a month                             | Opus 5.5 at 2x and Sonnet 5.5 at 1.3x the Auto credit rate | 1,000 to 10,000 credits a month; credits measure work, not prompts                          |
| OpenCode Go                                  | $10 or $40 a month                              | Haiku 5.5 only                                             | $15 or $60 of Haiku usage a month, with five-hour and weekly sub-limits                     |
| Cursor on the new X tiers                    | From $30 a month (Premium)                      | Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5                 | Not included: billed at API list price from the first request                               |

Sources, read 20 October 2026: [Antigravity models](https://www.antigravity.google/docs/models/) and [plans](https://www.antigravity.google/docs/plans), [Google AI plans](https://gemini.google/subscriptions/); [Copilot billing](https://docs.github.com/en/copilot/concepts/billing-and-usage/individuals/billing), [supported models](https://docs.github.com/en/copilot/reference/ai-models/supported-models) and [model prices](https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing); [Kiro pricing](https://kiro.dev/pricing/) and [models](https://kiro.dev/docs/models/); [OpenCode Go](https://opencode.ai/v2/docs/console/go/); [Cursor models and pricing](https://cursor.com/docs/models-and-pricing).

* **Antigravity** is Google's agent harness (an app, a CLI and IDE extensions), so this is Claude inside Google's tools, not Claude Code. The 5.5 models are "\[a]vailable on Google AI Pro for non-trial subscriptions only". Sonnet 4.6 and Opus 4.6, the Claude models Free and AI Plus accounts can pick, are marked "Will be removed on November 2, 2026"; on the current matrix that leaves those tiers, and Pro trials, with no Claude model after that date ([Antigravity models](https://www.antigravity.google/docs/models/)). Google keeps non-Gemini models "on a separate fixed rate limit" because of capacity ([Antigravity plan changes](https://www.antigravity.google/blog/changes-to-antigravity-plans)).
* **Copilot's** allowance is base credits equal to the price plus a "flex allotment" that GitHub describes as variable; included credits reset at 00:00 UTC on the first of each month and do not roll over ([Copilot billing](https://docs.github.com/en/copilot/concepts/billing-and-usage/individuals/billing)). Its Claude token rates match Anthropic's list prices ([model prices](https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing)), so Max at $100 buys about $200 of Claude at list price while the flex allotment stays where it is.
* **X and xAI plans** do not include Claude. The X Subscription's shared pool covers Grok and Cursor's own models; third-party models such as Claude "aren't included in your plan" and are "billed at API list prices from the first use" ([X help](https://help.x.com/en/using-x/x-subscription)). Older Cursor subscriptions have different terms ([legacy Cursor plans](https://cursor.com/docs/account/pricing/legacy-cursor-plans)).

![Google Antigravity's reasoning model table: Claude Sonnet 5.5 and Claude Opus 5.5 (thinking) are not available on Free and Google AI Plus, available on Google AI Pro for non-trial subscriptions, Google AI Ultra and Enterprise; Claude Sonnet 4.6 and Opus 4.6 (thinking) are available on Free, Google AI Plus and Google AI Pro but not Ultra or Enterprise, and are marked as being removed on November 2, 2026.](https://cdn.moda.app/voyager/blog/claude-usage-limits/fig-antigravity-eaba5e6196d8f71f.webp)

Which Claude models each Google AI plan gets in Antigravity, from Google's docs, read signed out on 20 October 2026. The 4.6 models leave on 2 November.

## How much a plan holds in a month

Anthropic does not say; its pricing FAQ says "there's no fixed message count" ([pricing](https://claude.com/pricing)). SemiAnalysis has estimated it for heavily cached coding work on Opus, at API prices:

| Plan    | Modelled full month on Opus 5.5, at API prices | Relative to Pro |
| ------- | ---------------------------------------------- | --------------- |
| Pro     | $1,178                                         | 1x              |
| Max 5x  | $5,725                                         | About 4.9x      |
| Max 20x | $11,726                                        | About 10x       |

Source: [SemiAnalysis](https://newsletter.semianalysis.com/p/anthropic-subscriptions-offer-5x), 5 October 2026, which moved the meter with one token type at a time and modelled a coding workload of 96.6% cache reads; not affiliated with Anthropic.

So the monthly room grows much less than the 20x name, which describes the five-hour session; for this workload Max 20x holds roughly twice Max 5x's. These are estimates for one modelled kind of coding work, not quotas, and not money saved: an allowance is worth its API price only in weeks you would otherwise pay for that usage.

## How the limits changed in 2026

If Claude Code seems to run out sooner than it did in the summer, that is partly real: against August, a Claude Code week now holds about 17% less; against April, 25% more, and a session 140% more. Anthropic's announcements:

* **6 May:** Claude Code's five-hour limits doubled on Pro, Max, Team and seat-based Enterprise, and the peak-hours reduction on Claude Code ended for Pro and Max ([Anthropic](https://www.anthropic.com/news/higher-limits-spacex)). The announcement covers Claude Code, not chat.
* **13 May to 13 September:** Claude Code's weekly limits were 50% higher as a promotion ([promotion terms](https://support.claude.com/en/articles/15910845-claude-code-may-august-2026-weekly-limits-promotion)).
* **20 July:** Fable moved to usage credits on Pro and Team Standard and to half the weekly limit on Max and Team Premium ([Fable on your plan](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan), [@claudeai](https://x.com/claudeai/status/2078302415804379218)).
* **14 September:** weekly limits in Claude Code settled at 25% above their pre-promotion level, which Anthropic said "works out to a 17% reduction in weekly limits on Claude Code" against the promotion ([@ClaudeDevs](https://x.com/ClaudeDevs/status/2093742322525810912)).
* **22 September:** with Opus 5.5, Claude Code's five-hour session limits rose 20%, and Pro, Max and Team users got a free reset ([@ClaudeDevs](https://x.com/ClaudeDevs/status/2102438800836489554)).
* **25 September:** Claude Code began finishing the step in progress at a five-hour limit, from "a small, fixed allowance pulled from your weekly limit" ([@ClaudeDevs](https://x.com/ClaudeDevs/status/2103561342057943314)).
* **7 October:** Max and Team gained monthly API credits ([release notes](https://support.claude.com/en/articles/12138966-release-notes)).

![Step chart of Claude Code limits in 2026 indexed to their level before 6 May: the five-hour limit rises from 100 to 200 on 6 May and to 240 on 22 September; the weekly limit rises from 100 to 150 on 13 May for a promotion, then settles at 125 on 14 September, 17% below the promotion.](https://cdn.moda.app/voyager/blog/claude-usage-limits/hero-71ee6aa5e1ef76fd.webp)

Claude Code's five-hour and weekly limits on Pro, Max and Team, as Anthropic announced the changes, indexed to their level before 6 May 2026. Chat and Cowork limits are not shown.

## Frequently asked questions

**Does Claude have a daily limit?** No daily allowance is documented. The limits are a five-hour session and, on paid plans, a week ([usage and length limits](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work)). If you run out early in the day, the message or Settings, then Usage, tells you which one it was and when it comes back.

**Does the five-hour limit mean five hours of work?** No. Five hours is how often the allowance refills, not how long it lasts. A Claude Code session with a large context, subagents or Opus at high effort can use up the allowance long before five hours pass ([Claude Code costs](https://code.claude.com/docs/en/costs#why-usage-climbs-in-a-long-session)).

**Do Claude chat and Claude Code share usage limits?** Yes, along with Cowork, the IDE extensions and the desktop and mobile apps ([usage and length limits](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work)). Anthropic has, however, raised Claude Code's limits separately several times in 2026.

**How many messages do you get on Claude Pro?** Anthropic does not publish a number, because usage depends on conversation length, attachments, the model, effort and tools: "there's no fixed message count" ([pricing](https://claude.com/pricing)). Short questions in fresh chats go much further than long conversations with files and research.

**Does Claude have a monthly limit?** No general monthly cap on included usage is published, as of 20 October 2026, though Anthropic reserves the right to add "weekly and monthly caps" at its discretion ([Pro plan](https://support.claude.com/en/articles/8325606-what-is-the-pro-plan)). The monthly limits that exist are the spend limit you set on usage credits, the cap on discounted bundles, and admin spend limits on Team and Enterprise.

**Does the Free plan have a weekly limit?** Anthropic documents only a five-hour session limit for Free, with messages varying "based on demand", and reserves the right to impose other limits ([Free plan](https://support.claude.com/en/articles/8114491-get-started-with-claude)). Free does not include Claude Code or Opus ([pricing](https://claude.com/pricing)).


## Voyager is a desktop AI agent for creative work

Describe what you want to make; Voyager plans it, uses your apps and tools, and keeps everything editable.

[Sign up](/sign-up)