Claude limits come in two layers. The session limit resets every five hours. The weekly limit covers all models and clears at a fixed time assigned to your account. Chat, Claude Desktop and Claude Code draw on one shared pool of tokens, so an hour in the terminal shrinks the quota you have left for conversations on the website.
Claude limits on Free, Pro and Max
Anthropic does not say how many messages you get. The help pages use multiples instead of numbers: Max 5x gives five times the Pro session limit, Max 20x gives twenty times. The figures below come from the pricing page and from Anthropic's help pages as of October 29, 2026.
| Plan | Price | Session limit | Weekly limit | Claude Code |
|---|---|---|---|---|
| Free | $0 | yes, message count not published | not described in the help pages | no |
| Pro | $20 per month, $17 when billed yearly | higher than Free | across all models | yes |
| Max 5x | $100 per month | 5 times Pro | across all models plus a separate one for Fable | yes |
| Max 20x | $200 per month | 20 times Pro | across all models plus a separate one for Fable | yes |
What you do matters more than how many messages you send. According to Anthropic's help pages, these things spend quota:
- the length of messages and the size of attachments;
- the conversation itself: the longer it gets, the more each next reply costs;
- tool calls and multi-step tasks;
- the chosen model and the effort level;
- artifacts.
Claude Code limits
Pro and Max include Claude Code in the terminal, in IDEs and in Claude Desktop. All three draw from the same quota as chat. Free does not list Claude Code among its features.
Two details are easy to miss. First, if ANTHROPIC_API_KEY is set in your environment and you approved it at launch, Claude Code runs on that key. The subscription sits idle and you pay API rates. Second, the API credits that come with Max: the monthly $100 or $200 does not apply inside Claude Code or the Claude apps. It covers only the API, the Agent SDK and Managed Agents.
The messages You've hit your session limit and You've hit your weekly limit apply to the whole account, so switching models with /model will not help. The exception is a message about the Opus or Sonnet limit: there, moving to a model from the other family works.
Where to check what is left
On the website and in Claude Desktop, open Settings > Usage. It shows bars for the session and weekly limits and the reset time. In Claude Code the /usage command shows the same bars and adds a breakdown of how much quota went to skills, subagents, plugins and individual MCP servers. The d and w keys switch the period between a day and a week. The Session line with dollar amounts on that screen is meant for API users, so subscribers can ignore it.
As you near a ceiling, Claude warns you: "Approaching session limit" or "Approaching weekly limit". Once the limit is used up, the message gives the reset time.
What to do when the limit runs out
There are four options, from free to paid.
- Wait. Since version 2.1.234, Claude Code can wait for the reset on its own and carry on with the interrupted task. You pick the mode with
/rate-limit-options. - Claim a limit reset. Anthropic hands out one-off resets from time to time: the "Reset for free" button sits in
Settings > Usageon the website and in Claude Desktop. Claude Code itself and the mobile app have no such button yet, but the limits are shared, so after a reset on the website they clear there too. - Turn on usage credits on Pro or Max. Requests then switch to payment at standard API rates. You set the prepayment and the monthly spending cap in
Settings > Usage, and from Claude Code the/usage-creditscommand opens that section. - Move to a higher plan. The limits of the new plan apply immediately.
Codex works in a similar way: five hours plus a week. How it plays out there is covered in Codex limits.
How to stay under the limit
Legitimate tricks cut consumption rather than hide it. Everything below appears in the Claude Code cost documentation.
- When you switch tasks, clear the context with
/clear. The old conversation rides along with every next request and adds nothing. - Do not leave a session of hundreds of thousands of tokens open overnight. The first message after a long pause misses the cache and recomputes the whole context. Before a break, compress the dialogue with
/compactand tell it what to keep. - Hand routine work to Sonnet and keep Opus for architecture and hard reasoning.
- Lower the effort level with
/efforton simple tasks: thinking is billed as output tokens. - Disable unused MCP servers with
/mcp, and keepCLAUDE.mdunder 200 lines. - Phrase the task precisely. "Add validation to the login function in auth.ts" costs less than "improve the project". For big tasks, switch on planning mode with Shift+Tab.
- In chat, put reference documents into Projects: rereading from the cache is cheaper, as long as the cache has not expired after idle time.
More tricks, with numbers, are collected in How to cut your Claude Code bill threefold and Prompt caching.
What we do not recommend: schemes with several accounts and other people's subscriptions. The vendor has its own rules, and a mistake costs you the account along with its history. We wrote separately about bans, their causes and appeals: VPN for Claude and account bans.
Token limits: context and API
In searches about limits, "tokens" mean two different things. The first is the context window. On paid plans it reaches 1M tokens for the newest models and 500K or 200K for the rest, and part of the window is reserved for the reply.
The second is API limits, and they work differently from a subscription. Anthropic's documentation sorts organizations into tiers, each with its own monthly spending cap and separate limits on requests and tokens per minute.
| API tier | Monthly spending cap | Opus 5.5: requests per minute | Input tokens per minute | Output tokens per minute |
|---|---|---|---|---|
| Start | $500 | 1 000 | 2 000 000 | 400 000 |
| Build | $1 000 | 5 000 | 5 000 000 | 1 000 000 |
| Scale | $200 000 | 10 000 | 10 000 000 | 2 000 000 |
The table shows the standard tier limits, and new organizations may begin on the Evaluation tier, where the ceilings are lower. For most models, cached input tokens do not count toward the input limit, so prompt caching raises your real throughput without a tier upgrade request.
Claude through the API: pay per token instead of windows
Subscription limits hit hardest for people who sit in Claude Code six hours straight. The API works differently: you pay for tokens, and there are no five-hour windows and no weekly cap. codexkey sells exactly this kind of access from Russia: one sk_cdx_... key, payment by a Russian bank card, through SBP (the Russian Faster Payments System) or in cryptocurrency. codexkey does not sell Claude Pro and Max subscriptions and does not give access to claude.ai. It is API only.
Prices in USD per 1M tokens, codexkey price list as of October 29, 2026. Current prices are on the models page:
| Model | ID | Input | Output | Cache read |
|---|---|---|---|---|
| Claude Opus 5.5 | claude-opus-5-5 | $0.86 | $4.32 | $0.10 |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | $0.52 | $2.60 | $0.06 |
| Claude Haiku 5.5 | claude-haiku-5.5 | $0.08 | $0.32 | $0.02 |
| Claude Fable 5.1 | claude-fable-5-1 | $7.48 | $37.44 | $0.76 |
Claude Code connects through ~/.claude/settings.json. The config comes from the Claude Code guide, and the address is written without /v1:
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.codexkey.online",
"ANTHROPIC_AUTH_TOKEN": "sk_cdx_YOUR_KEY",
"ANTHROPIC_MODEL": "claude-sonnet-5[1m]",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-5-5[1m]",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-5[1m]",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-sonnet-5",
"CLAUDE_CODE_SUBAGENT_MODEL": "claude-sonnet-5"
}
}
Sonnet 5 costs the same with us as Sonnet 5.5. The ANTHROPIC_DEFAULT_* lines are required: without them Claude Code asks for its own default models and fails with a model selection error.
The gateway has no windows, but the provider's ceilings remain. Under overload you get HTTP 429, so retry with exponential backoff. Specific RPM and TPM figures are not published in the rate limits guide, and you cannot configure them for a key.
An honest caveat: for daily multi-hour work with Opus, a Max subscription often comes out cheaper than tokens. We would not switch to the API to save money until you have measured your own consumption with /usage. The rate calculations are in Claude subscription prices: Free, Pro, Max.
FAQ
What are the Claude limits?
Paid plans have two: a session limit that resets every five hours, and a weekly limit across all models. Max has a separate weekly limit on Fable as well. Anthropic does not publish an exact message count; it depends on conversation length, the model and the effort level.
How many messages can I send on Claude Pro?
There is no fixed number. Pro gives a larger session limit than Free, and Max 5x and Max 20x are multiples of the Pro session. One long conversation with attachments burns quota faster than ten short questions.
When does the Claude weekly limit reset?
At a fixed time that Anthropic assigns to your account. Settings > Usage shows the date and hour. The session limit runs in its own five-hour cycle meanwhile.
Are the limits shared between Claude Code and chat?
Yes. Chat on claude.ai, Claude Desktop and Claude Code, IDE extensions included, draw from one quota. There is one exception: if Claude Code runs on ANTHROPIC_API_KEY, it pays API rates and leaves the subscription quota alone.
How can I raise my Claude limits?
Three approaches work: move to Max 5x or Max 20x, turn on usage credits on Pro or Max, or switch to paying for tokens through the API. Savings habits such as /clear, /compact and choosing Sonnet over Opus also stretch your quota at no extra cost.
Is there a Claude without limits?
No. Paying per token removes the subscription windows, but API limits stay: at Anthropic these are tiers with requests-per-minute and tokens-per-minute limits and a monthly spending cap, and at the gateway it is HTTP 429 under overload. On the Free plan, Claude Code is not available at all.
What are the limits on free Claude?
The Free plan has a usage limit, but Anthropic does not publish the message count: the pricing page only says that restrictions apply. Claude Code is not part of the free plan. Which free routes to a coding agent remain is covered in Claude Code for free.