Orpheus
Token adoption in teams

Are you using the Claude capacity you pay for?

Short answer

Check each seat's share of its weekly limit used by the reset, averaged over four weeks. A $100 Max plan that ends its weeks at 30% leaves most of the week's allowance unused. Raise it with real work: more agents in parallel, long tasks in the background, separate worktrees. Move or resize seats nobody uses.

On this page

What utilization means

Claude subscriptions do not sell tokens. They sell an allowance that resets: Max is described as “5x Pro usage every 5 hours and every week”,1 and each account has a fixed weekly reset time shown in Settings > Usage.2 So the useful question is not how many tokens a seat used, but how much of each week's allowance it used before the week reset.

Utilization = the 7-day percentage shown just before the weekly reset, per seat, averaged over the last four resets.

The number is in Settings > Usage on claude.ai, and in the statusline data Claude Code passes to scripts as rate_limits.seven_day.used_percentage with resets_at, for Pro, Max and Ultra logins.3 Usage in Claude chat, Claude Code and Claude Desktop counts toward the same limit,2 so the percentage covers everything the person did on that account. A logging script is in How do you measure Claude Code adoption across a team?

Plans as of November 2026: Pro $20 a month, Max $100, Ultra $200 (formerly Max 5x and Max 20x, renamed on October 30, 2026).14 Team plans run 2 to 150 seats: standard seats are $25 a month ($20 billed annually) and premium seats $125 ($100 annually), with “5x more usage than standard seats”.1

Worked example with the dollar math

Illustrative numbers. The percentages in this section are made up to show the method. They are not data from any company or from Orpheus users.

One person on Max

Four weekly resets end at 25%, 35%, 30% and 30%. Utilization is 30%. A rough reading of the $100: about $30 a month went to capacity that was used and about $70 to capacity that sat idle. This is a way to size the gap, not a refund value, because the plan price is not a per-token meter.

A team of six on Team monthly seats

Illustrative: share of weekly limit used at reset, four-week average
SeatTypePrice a monthUsed at resetPrice × unused share
1Premium$12580%$25
2Premium$12512%$110
3Standard$25100%, limit hit 3 of 4 weeks$0
4Standard$2540%$15
5Standard$250%$25
6Standard$250%$25
Total$35043% weighted$200
Illustrative example by Orpheus. Seat prices from the Claude pricing page,1 checked .

Weighted utilization is $150 of used capacity out of $350, about 43%. What to do with each row:

  • Seat 2 uses 12% of a premium seat. A premium seat has 5 times a standard one, so 12% of it is about 60% of a standard seat. Moving to standard saves $100 a month with room to spare.
  • Seat 3 runs out most weeks. Give it the premium seat that seat 2 no longer needs. Spend stays at $350, and the capacity moves to the person who uses it.
  • Seats 5 and 6 are idle. Ask why before removing them: no onboarding, a blocked repo, or no task that fits. Often a pairing session fixes it.
  • Seats 1 and 4 are fine. A seat does not need to reach 100%; a margin keeps people from stalling before the reset.

When a smaller seat would do

Because Max is described as 5 times Pro, and a premium Team seat as 5 times a standard one,1 a rough rule follows:

Rough resize rule from the published multipliers
SeatFour-week utilizationReading
Max ($100)Under 20% every weekWould likely have fit in Pro ($20). Check the 5-hour window too: bursts can still need Max.
Team premium ($125)Under 20% every weekWould likely have fit in a standard seat ($25)
Any seatHits 100% before the reset most weeksSize up, or turn on usage credits with a cap
Derived from Anthropic's plan descriptions, which are not exact token figures. Ultra is left out because its published multiplier has changed.

How to raise use, legitimately

Low utilization usually means agents wait on people, or people only run one agent at a time. Ways to put real work on the capacity:

  1. Run agents in parallel on separate worktrees

    One git worktree and one tmux session per task lets two or three agents work without stepping on each other's files. See How do you run multiple Claude Code agents?

    # a second agent on its own branch and folder git worktree add ../app-tests -b add-tests tmux new -s app-tests -c ../app-tests
  2. Hand long tasks to the background

    Test coverage, dependency upgrades, migrations and documentation are work an agent can do while you are in meetings. Queue them at the end of the day rather than leaving the seat idle overnight.

  3. Cut the wait on humans

    An agent stuck on a permission prompt uses nothing. Answer from your phone, or allow safe commands in project settings so agents keep moving.

What not to do

  • Don't run busywork to move the number. Utilization is a cost signal, not a goal. Amazon scrapped a usage leaderboard after people gamed it, and compute costs went up.5
  • Don't share a seat. Anthropic's Consumer Terms say you “may not make your Account available to anyone else.”6 Move the seat to the person instead.
  • Don't create accounts to stretch limits. The Usage Policy forbids creating multiple accounts “to avoid detection, circumvent product guardrails”.7 This is not legal advice; see the terms guide.
  • Remember why weekly limits exist. Anthropic added them in August 2025 after some users ran Claude Code continuously in the background, around the clock.8 Limits can change at Anthropic's discretion,2 so plan on real work, not on maximizing the meter.

Common questions

Does unused weekly capacity roll over?

No published rollover exists. Each account has a fixed weekly reset time, and the allowance starts again at the reset, so capacity left at the reset is gone.

What utilization should we aim for?

There is no published benchmark. A seat that ends most weeks between about half and most of its limit, without hitting it early, is using what it pays for while leaving a margin. Set your own range from four weeks of data.

Is it allowed to keep agents busy to use the capacity?

Doing real work on your own subscription is ordinary use. Running filler tasks to push the number up wastes effort and shared capacity, and sharing an account with another person is against Anthropic's Consumer Terms.

Where do I see my weekly percentage?

In Settings > Usage on claude.ai, and in the rate_limits fields of Claude Code's statusline data for Pro, Max and Ultra logins.

Sources

  1. Plans and pricingClaude, undatedChecked
  2. How do usage and length limits work?Claude Help Center, undatedChecked
  3. Customize your status lineClaude Code docs, undatedChecked
  4. Anthropic renames Claude Max plans and drops Ultra's five-hour limitThe Next Web, Nov 1, 2026Checked
  5. The tokenmaxxing era is over. Now companies are 'tokenminimizing'The Next Web, Jun 17, 2026Checked
  6. Consumer Terms of ServiceAnthropic, effective Oct 8, 2025Checked
  7. Usage PolicyAnthropic, effective Nov 12, 2026Checked
  8. Anthropic unveils new rate limits to curb Claude Code power usersTechCrunch, Jul 28, 2025Checked