Frontier plans
Frontier plan
iqshard · Same price
Claude Pro
$20 pcm
- The baseline. Session limit resets every five hours. The size of it is not published.
- A weekly limit on top of that, also not published.
- Past either one, the plan bills you at retail to carry on.
- Then it stops until you pay or the window rolls.
iqshard · 1 × Pro Share
$20 pcm
- 1,297 requests a month at P1
- 6,486 requests a month at P2
- Unlimited at P3
- No meter, and no second bill.
Every figure here is published and worked out on the page. None of the ones opposite are.
Claude Max 5×
$100 pcm
- Five times Pro. Session limit resets every five hours. The size of it is not published.
- A weekly limit on top of that, also not published.
- Past either one, the plan bills you at retail to carry on.
- Then it stops until you pay or the window rolls.
iqshard · 5 × Pro Shares
$100 pcm
- 6,486 requests a month at P1
- 32,432 requests a month at P2
- Unlimited at P3
- No meter, and no second bill.
Every figure here is published and worked out on the page. None of the ones opposite are.
Claude Max 20×
$200 pcm
- Twenty times Pro. Session limit resets every five hours. The size of it is not published.
- A weekly limit on top of that, also not published.
- Past either one, the plan bills you at retail to carry on.
- Then it stops until you pay or the window rolls.
iqshard · 10 × Pro Shares
$200 pcm
- 12,973 requests a month at P1
- 64,865 requests a month at P2
- Unlimited at P3
- No meter, and no second bill.
Every figure here is published and worked out on the page. None of the ones opposite are.
Every row is a frontier plan against the Share bundle that costs exactly the same. The prices line up to the dollar. What does not line up is that one side of each row can tell you what you get.
Requests here are one coding turn: 52,000 cached tokens, 700 new, 150 out. A long-context agent turn runs several times heavier than that, so put your own shape in and the counts move with it. Which model suits your work is a separate question from how much of it you are allowed to send.
We went looking for the number
“How much do I get for $20?” is the only question anyone asks about a coding plan. It should have an answer. We spent a while trying to find one.
1Ask Anthropic
The pricing page says Max gives you “5x or 20x more usage than Pro.” More than what is not stated. Two help-centre articles cover usage limits and neither contains a figure; one explains that your usage depends on “the length and complexity of your conversations, the features you use, which Claude model you’re chatting with, and the effort level you’ve selected.”
2Measure the meter
Claude Code shows a session meter as a percentage. Read it twice and the difference should give you the whole. Here are three readings of one session, minutes apart.
| Between readings | Meter | Cost | Requests | So a full window is |
|---|---|---|---|---|
| First to second | +4 points | $2.74 | 10 | $68.50 · 250 requests |
| Second to third | +4 points | $1.02 | 3 | $25.50 · 75 requests |
Same meter, same session, the same four points. Two answers a factor of 2.7 apart. Whatever it is counting, it is not dollars, and it is not requests, and it is not tokens either — the cache-read figures move by the same ratio.
3Try the weekly meter instead
A week reading 94% used, against 1.5 billion tokens processed, puts the limit near 1.6 billion. Price that at the rates the same session was billed at and a $20 plan is handing over something like $3,700 of inference a month. Nobody reports a return in that neighbourhood, and the client itself says why: usage is weighted by how long your context has grown, not by how many tokens went past. Longer sessions cost more even when everything is cached.
4Take somebody else's number
Three published estimates, three different units, no agreement. One counts Sonnet-hours a week. One counts tokens per five-hour window. One counts API-equivalent dollars. The token one puts a five-hour window between 44,000 and 90,000 tokens; a single measured window read 105 million, and its output alone was 375,000. They are not describing the same quantity.
Four attempts, no number. And a number would have gone stale anyway: the five-hour limits doubled on 6 May 2026, and a promotion running to 31 August lifted the weekly limit by half again. Anthropic’s own client labels the usage it shows you approximate, and notes it cannot see your other devices.
None of which is carelessness
- A limit you do not publish is a limit you can move, and moving it is how you keep a plan viable when demand changes.
- The cost lands on the person trying to plan a month of work around it.
A Share is the other trade
- It is published. Bucket maximums and refill rates, with the arithmetic on the page.
- You can check it before you buy. Work out your own entitlement against your own workload.
- Nothing stops. Past your entitlement the work still goes out, at P3 — no bucket, no charge, no window.
- That is the trade. Sometimes you wait. In exchange the number stops moving.
See what is available
Pledge while a server is funding, buy once it is confirmed or live, or join the waiting list for the next opening.