How many tokens do you get with Claude Max?

Anthropic does not publish a number. In a Claude Code workload a Max 20x week measured about 1.9 billion tokens and a Max 5x week at least 1.3 billion, but 96% were cache reads; output was about 11 million tokens a week on Max 20x.

Quick answer: Anthropic doesn’t publish a token limit for Claude Pro or Max. Measured from real Claude Code logs, a Claude Max 20x week holds about 2 billion tokens and a Max 5x week at least 1.3 billion....

How many tokens do you get with Claude Pro and Max? We counted them

Q: How many tokens do you get with Claude Pro?

Anthropic does not publish a Pro token limit. Scaling a measured Max 5x session (about 39 million tokens) by Anthropic’s 5x multiplier projects a Pro 5-hour session at about 8 million tokens in a coding-agent workload.

Q: When does the Claude token limit reset?

The session limit resets every five hours. The weekly limit resets at a fixed time each week that Anthropic assigns to your account.

Updated September 27, 2026 · first published September 27, 2026

Anthropic doesn’t publish a token limit for Claude Pro or Max. Measured from real Claude Code logs, a Claude Max 20x week holds about 2 billion tokens and a Max 5x week at least 1.3 billion. That sounds enormous, but 96% of those tokens are cache reads, the agent re-reading its own context. Only about 11 million are output tokens, the text Claude actually writes.

This page answers one question from our 60-day study of Claude Max against ChatGPT Pro and Codex. Every figure is priced at the Anthropic API list price in force on the day of each request, then converted back to tokens using the August 2026 Opus mix: about 1.05 million tokens per dollar.

Does Anthropic publish a token limit for Claude Pro or Max?

No. Anthropic describes Max as 5x or 20x Pro’s usage per session and says the session limit resets every five hours, with a separate weekly limit that resets at a fixed time assigned to your account. It meters both in its own units, which weigh pricier models more heavily, so no single token number would be true for every model anyway.

How many tokens do you get per week on Claude Max?

About 1.9 billion on Max 20x and at least 1.3 billion on Max 5x, in a Claude Code workload.

Plan	Price / month	API $ per week	All tokens per week	Output tokens per week	Basis
Claude Max 20x	$200	~$1,860	~1.9B	~11M	Measured (weekly cap hit twice)
Claude Max 5x	$100	≥ $1,217	≥ 1.3B	≥ 7M	Lower bound (cap never hit)
Claude Pro	$20	Not measured			Weekly cap not published

Max 20x’s range runs from about 1.8 to 2.3 billion tokens, depending on how much of the week goes to pricier models. The weekly-limit breakdown explains why the $200 plan’s week is only 1.5 to 2 times the $100 plan’s.

How many tokens do you get per 5-hour session?

About 39 million tokens on Max 5x, or roughly $37 of API usage. That is the median at which the 5-hour limit stopped our user on Max 5x, twelve times in September 2026. Scaled by Anthropic’s own multipliers, a Pro session projects to about 8 million tokens and a Max 20x session to about 155 million. Those two are projections, not measurements.

Why are almost all of those tokens cache reads?

Because a coding agent re-sends its whole working context on every step. Anthropic bills that repeat as a cache read at a tenth of the input price, and across 7.3 billion tokens that tenth still came to half the bill.

7.0 billion cache-read tokens (96% of volume) made 50% of the $6,986 Opus 5 bill. Output tokens, the figure most cost models are built on, were 0.6% of volume and 15% of cost. Cache writes at the 1-hour rate ($10/M) are the hidden second line.

So “how many tokens do I get?” depends on how you use Claude. A chat user who sends short messages with no long context gets far fewer total tokens from the same limit, but a much larger share of them are output.

How do you convert Claude tokens to dollars?

Multiply each token type by its own rate. On Opus 5 in August 2026, those rates were $0.50 per million for cache reads, $10 for 1-hour cache writes, $25 for output and $5 for fresh input. Our user’s month came out as follows.

Opus 5, August 2026	Tokens	Cost
Cache reads	7,021M	$3,510
Cache writes, 1-hour	243.7M	$2,437
Output	41.5M	$1,038
Fresh input	0.22M	$1
Total	7,307M	$6,986

Claude Code writes every request, with its token counts, to the JSONL files under ~/.claude/projects, so you can run the same sum on your own usage. To check and cut it from inside Claude Code, see how to reduce Claude Code token usage.

When does the Claude token limit reset?

Every five hours for the session limit, and once a week at a fixed time for the weekly limit. The weekly reset time is set per account, not the same day for everyone. ChatGPT’s Codex works differently: its weekly window starts from your first request, as our Codex limits breakdown explains.

Sources

Want this applied to your own LLM spend? FinOps LLM runs a free audit of your AI costs and shows where the savings are. Book free audit →

Back to research