Robin Hood · AI limits watch
How much of your weekly limit does one Claude or Codex session use?
Your plan shows two gauges: the 5-hour session and the week. Nobody publishes how they relate. Robin Hood measures it on real subscriptions: each time a session gauge moves, how far the weekly gauge moved with it. No token counts, no guesswork, just what one session is worth in your week, and whether that changed.
Updated Sep 25, 2026, 12:59 AM · last 30 days compared with the 90 days before
Sessions measured
0
Accounts
0
Members
0
The first measurements are coming in
No plan has reached three measured accounts yet, so there is nothing to publish. Every AgentsRoom desktop that shares usage data adds a sample at the end of each 5-hour session: the index fills itself as people work.
Download AgentsRoomHow it works
The whole method fits in eight lines. Nothing is estimated from tokens or prices.
- 1
Two gauges, one account
Claude, Codex and a few other CLIs publish a session gauge (5 hours) and a weekly gauge for the same subscription. AgentsRoom already reads both to show your usage.
- 2
Same window, both deltas
Between two readings taken in the same session window and the same week, the desktop notes how many points each gauge moved. When the session window closes, that becomes one sample.
- 3
A ratio of sums
The index divides the total weekly points by the total session points. A short 6-point session does not count as much as a 90-point one, as it would with an average of ratios.
- 4
{recent} days vs the {baseline} before
Each card shows the last 30 days. The trend compares them with the 90 days before, and the curve shows the published weeks one by one.
- 5
A change means {threshold}% or more
An alert is raised only when the recent ratio differs by at least 20% from the previous period, and only when both periods are published. Small drifts are noise.
- 6
At least {min} accounts
No number is shown until 3 different accounts have been measured on that plan. Below that, the card keeps its counters and says why the figure is hidden.
- 7
Shape only
Each sample carries the provider, the plan, the gauge, two numbers of points and a salted hash of the account for deduplication. Never an email, a prompt, a project or a token.
- 8
Your choice
Only desktops that keep "Share usage data" on contribute. Turn it off in the settings and your measurements stay on your machine, where you still see your own ratio.
Want a warning before you hit a limit, or to see which models the community runs? Usage limit alerts · LLM usage stats
What the ratio does and does not say
A useful number, as long as it is read for what it is.
It tells you
- How much of your week one full session costs, and so how many full sessions fit in a week on your plan.
- How plans compare with each other on the same provider, measured the same way on real accounts.
- When that weight moves suddenly, which is what a quiet change of limits looks like from the outside.
It does not tell you
- An absolute limit. A session that weighs more can mean a smaller week or a bigger session: the ratio cannot tell which side moved.
- How much work fits in a session. That depends on the model, the context size and the task, not on the ratio.
- A number of tokens or dollars. Subscriptions meter usage in percentages of their own windows, and that is what is measured here.
Questions about session and weekly limits
How much of my weekly limit does one Claude session use?
It depends on your plan and it changes over time, which is why this page measures it instead of guessing. Each card above shows, for one plan, what one full 5-hour session costs in percent of the weekly limit, measured over the last 30 days on real accounts. A session you only use halfway costs about half of that.
Did Claude reduce its weekly limits?
Providers rarely announce every adjustment. When the weight of a session in the week moves by 20% or more compared with the previous three months, this page shows it at the top. A higher weight means the week holds fewer full sessions than before, whether the week got smaller or the session got bigger.
How does the Codex 5-hour limit compare with its weekly limit?
Codex publishes the same two gauges as Claude: a 5-hour window and a weekly one. The Codex cards show how many points of the week a full 5-hour window costs on Plus, Pro and the other plans, measured the same way, so both providers can be compared directly.
How many Claude sessions per week do I get on Max 5x or Max 20x?
Divide 100 by the share of the week one session uses: a session worth 10% of the week gives about 10 full sessions. Each card does that division for you, per plan. Most weeks are made of partial sessions, so treat it as a ceiling, not a schedule.
Why are token counts meaningless for subscription limits?
A subscription does not give you a token balance. It gives you two gauges in percent, and what moves them depends on the model, the context and the load on the provider side. The same number of tokens can cost very different amounts of your week, which is why this index measures the gauges themselves.
Where does the data come from, and is it private?
From AgentsRoom desktops whose users keep "Share usage data" on. They send the plan, the gauge and two numbers of points per session window, with a salted hash of the account used only to avoid counting the same subscription twice. No number is published until at least three different accounts are behind it.
See your own ratio, live
AgentsRoom reads the session and weekly gauges of every account you sign into, shows what your own sessions cost in your week, and warns you before a limit stops an agent.
Download AgentsRoom