Claude enforces three usage caps at account level: a rolling 5-hour session limit, a weekly limit across all models, and a second weekly limit scoped to one model family. They count simultaneously, and the same allowance covers Claude chat, Claude Code, and Cowork. There is no fixed message count, because a message costs whatever its context, model, and tool use cost. Check yours in Settings then Usage on claude.ai, or with /usage inside Claude Code.
- Limits are per account, not per product. Chat and Claude Code draw on one pool.
- Three caps: a 5-hour session limit, a weekly limit, and a model-scoped weekly limit.
- There is no message count. Context size and model choice drive consumption.
- Anthropic doubled the 5-hour Claude Code limits on 6 May 2026 and dropped the peak-hours reduction.
- Only a model-scoped limit is fixed by
/model. The other two ignore which model you pick.
Three limits, one account
Anthropic does not sell you a number of messages. It sells you an allowance, and three separate meters draw on it at the same time. Every message you send in Claude chat, every turn an agent takes in Claude Code, and every Cowork run counts against the same pool, which is why a heavy morning in the terminal can leave you rate limited in the browser an hour later.
The three caps, as of August 2026.
| Cap | Window | Covers | The error text |
|---|---|---|---|
| Session limit | Rolling 5 hours | All models | You've hit your session limit · resets 3:45pm |
| Weekly limit | One week, fixed reset | All models | You've hit your weekly limit · resets Mon 12:00am |
| Model limit | Rolling, model-scoped | One model family | You've hit your Opus limit · resets 3:45pm |
They count at the same time
Usage draws down every applicable meter simultaneously. A single intense afternoon can exhaust a weekly allowance while the 5-hour bar still shows room, which is why people report being blocked on Tuesday having felt fine on Monday. The weekly limit is a budget for the week, not a ceiling you only meet at the end of one.
Why there is no message count
The question people actually type is "how many messages do I get". There is no answer, and the reason is structural rather than evasive: two messages of the same length can differ in cost by two orders of magnitude depending on what is already in the conversation.
What Anthropic lists as driving the cost of one message.
| Factor | Effect on your allowance |
|---|---|
| Conversation length so far | Usually the largest. The whole thread is re-sent every turn |
| Model choice | Several times the draw between the cheapest and most capable |
| Effort level | Reasoning tokens are billed as output |
| Message length | Direct, and usually the smallest factor |
| File attachments | Charged once on upload, then re-sent with every later turn |
| Tool use (research, web search) | Each tool result is another request carrying the whole thread |
| Artifacts | Written, then carried in context |
This is why a one-line question at 5pm in a session that has been open since 9am can cost more allowance than fifty fresh questions would have. Claude Code flags exactly this: on a paid plan, the /usage breakdown raises a behavior flag for long context or cache misses when either accounts for 10% or more of your recent usage.
/clear # between unrelated tasks. Old context costs you on every message.
/model # match the model to the job rather than defaulting upWhat each plan buys
Anthropic publishes plan capacity as multipliers rather than token counts, so the honest table is a table of multipliers. Prices are as of August 2026 and are quoted for monthly billing, with the annual price in brackets where it differs.
| Plan | Price | Session allowance | Claude Code |
|---|---|---|---|
| Free | $0 | Baseline, and varies with demand | Not included |
| Pro | $20/mo ($17 annual) | At least 5x free per 5-hour session | Included |
| Max 5x | $100/mo | 5x Pro per session | Included |
| Max 20x | $200/mo | 20x Pro per session | Included |
| Team standard seat | $25/seat/mo ($20 annual) | Per-seat allowance | Included |
| Team premium seat | $125/seat/mo ($100 annual) | Larger per-seat allowance | Included |
| Enterprise | $20/seat plus usage at API rates | Per-seat allowance | Included |
On Team and Enterprise plans the seat allowance resets on the same rolling five-hour window and weekly window, and is shared with Claude chat and Cowork exactly as it is on personal plans. The difference is that an admin can turn on usage credits and set spend limits at organisation, group, or member level.
How to check what is left
On claude.ai: Settings, then Usage
Progress bars for the five-hour session limit with time remaining, plus the weekly limits and their reset times. This is the authoritative view because it is server-side and covers every device you use.
In Claude Code: /usage
/usage # plan usage bars, attribution, and behavior flags
# press d for the last 24 hours, w for the last 7 daysOn a Pro, Max, Team, or Enterprise plan the same screen breaks recent usage down by skill, subagent, plugin, and individual MCP server, each as a percentage of the total. That attribution is the fastest way to find an MCP server quietly costing you a fifth of your allowance.
Know what /usage cannot see
The breakdown figures are approximate and computed from the local session history on that machine. Work you did from a second laptop, from claude.ai, or from a phone is not in them. The plan bars themselves come from a server endpoint, so those do reflect everything.
The May 2026 doubling, and the other "double usage"
On 6 May 2026 Anthropic doubled the five-hour Claude Code rate limits for Pro, Max, Team, and seat-based Enterprise plans, and removed the peak-hours limit reduction for Pro and Max accounts. Both took effect the same day with no phase-in. Any guidance about Claude limits written before that date understates them, and any guidance describing your allowance shrinking during busy periods is now simply wrong.
- Doubled: the five-hour Claude Code limit, on Pro, Max, Team, and seat-based Enterprise.
- Removed: the peak-hours reduction on Claude Code for Pro and Max.
- Raised: API rate limits for Opus-class models, which is a separate system covered on the rate limit page.
The other thing people mean by double usage
The phrase also gets used for the feeling that one action is being counted twice. It usually is, and legitimately, for one of these reasons.
| What you did | Why it counted more than once |
|---|---|
| Sent one message | It drew on the session limit and the weekly limit |
| Used Opus | It drew on the model limit as well as both windows |
| Ran two sessions at once | One account, one pool, two draws |
| Used an agent team | Each teammate keeps its own context window and consumes until it exits |
| Ran a scheduled task | It fires on its interval and sends the full context each time, idle or not |
| Compacted a huge session | /compact reads the conversation it summarises, so it is itself a large request |
When you hit one
- Read which cap it names. Session and weekly ignore your model; a model-scoped limit does not.
- If it names a model, run
/modeland pick another. You keep working immediately. - If it names the session, the window is rolling, so headroom returns gradually rather than at a stated moment. Use the gap to read the diff.
- If it names the week, waiting is not a plan. Change model defaults, clear between tasks, or add capacity.
- Commit first, always. Being stopped is cheap. Being stopped mid-refactor with a half-applied change is not.
Buying past the limit
Usage credits let a Pro, Max 5x, or Max 20x subscriber keep working after the included allowance runs out. They are billed at standard API rates rather than a flat fee, you enable them under Settings then Usage with a payment method and an optional monthly cap, and there is a $2,000 daily redemption limit. Subscribers who bought through a mobile app store have to enable them on the web.
/usage-credits # Pro and Max: opens billing settings
# Team and Enterprise: asks your admin, once you confirmThe structural answer
If you routinely run two or three agents at once, one subscription is the wrong shape rather than the wrong size: parallel sessions divide one pool. Two Max 5x accounts cost exactly what one Max 20x costs and give you two independent sets of windows, which is a different product from twenty times one window.
Questions people ask
Three caps on one account allowance: a rolling 5-hour session limit, a weekly limit across all models, and a second weekly limit scoped to one model family. They apply simultaneously, and the same allowance covers Claude chat, Claude Code, and Cowork.
There is no fixed number. Consumption tracks the cost of each message, which depends on the conversation length so far, the model, the effort level, attachments, and tool use. A short question in a long thread can cost more than a long question in a fresh one.
The session limit is a rolling five-hour window, so capacity returns gradually as earlier usage ages out rather than at a fixed moment. The weekly limits reset at a set time each week, and the error message states it.
On claude.ai, open Settings then Usage for progress bars on the session and weekly limits. Inside Claude Code, run /usage for plan bars plus a breakdown by skill, subagent, plugin, and MCP server, with d and w toggling 24 hours and 7 days.
Yes. On Pro and Max, usage limits are shared across Claude and Claude Code, so all activity in both counts against the same allowance. Cowork draws on it too.
Yes, for Claude Code. On 6 May 2026 Anthropic doubled the five-hour Claude Code rate limits for Pro, Max, Team, and seat-based Enterprise plans, and removed the peak-hours limit reduction for Pro and Max.
The free plan has a session-based allowance that resets every five hours, with no published message count because it varies with demand. Claude Code is not included on free at all, so agent coding needs Pro or above, or an API key.
Only for a model-scoped limit such as the Opus limit, where /model gets you working again immediately. The session and weekly limits are shared across all models, so switching does nothing for them.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- Claude Code usage limit errors exact session, weekly, and model limit messages
- Higher usage limits for Claude 6 May 2026 doubling and peak-hours removal
- Manage usage credits for paid Claude plans eligibility, API-rate billing, $2,000 daily cap
- Claude plans and pricing plan prices as of August 2026