The core idea: you get one capacity pool, and each model spends from it at a different rate.
In one line: capacity is a pool of tokens measured in Haiku-equivalent; the model you pick (Haiku / Sonnet / Opus) decides how fast you spend it (1× / 3× / 5×); credit covers anything beyond the weekly bar.
The 3 models, from fastest to most powerful
Three Claude (Anthropic) models. Pick the right tool for the task:
Capacity: one pool, metered in Haiku-equivalent
Your capacity (the weekly and monthly bars) is metered in Haiku-equivalent tokens — one shared pool. Each model drains it weighted by its real cost: Haiku 1×, Sonnet 3×, Opus 5×.
So the same allowance gives roughly 5× more usage on Haiku than on Opus. Switching to Opus moves the bar 5× faster.
For the same capacity pool: how much each model lets you do.
What each plan includes
The included weekly capacity, per plan (the monthly one is the bigger envelope):
| Plan | Models | Capacity / week | Capacity / month |
|---|---|---|---|
| Starter | Haiku · Sonnet | 500k | 2M |
| Pro | + Opus | 1.5M (3×) | 3M |
| Business | + Opus | 7.5M (15×) | — |
Capacity expressed in Haiku-equivalent tokens. Pro and Business values are respectively 3× and 15× the Starter capacity.
When the weekly bar hits zero
Two options, never a hard stop:
1. Use credit — rechargeable credit (pay-as-you-go, also weighted 1× / 3× / 5×) takes over so you continue immediately.
2. Wait for the reset — the weekly bar resets every Monday; you resume at no extra cost.
Weekly vs monthly: the monthly bar is the bigger envelope; the weekly bar is the rate limit inside it (Claude-style). Each bar shows its own reset countdown.
In short
Capacity = a Haiku-equivalent token pool. The model you pick (Haiku / Sonnet / Opus) decides how fast you spend it (1× / 3× / 5×). Credit covers anything beyond the weekly bar.