A monthly allowance for people who leave a coding agent running. Same models, same caps, same receipts as pay-as-you-go, with the top up already done.
A few hours of agent work a day on a small repo.
Day to day development on a mid-sized repo.
All-day agent sessions, several repos, several people.
Your plan lands a credit allowance at the start of each period. It is spent on tokens at the same per-model rates as pay-as-you-go, so the prices on the models page are the prices you pay. There is no second currency to convert in your head.
Unused allowance does not roll over. If you run out, top up at the normal rate and carry on; nothing stops working and no request is throttled because of which tier you are on.
Every tier reaches every model, including the uncensored and abliterated catalogue. Every tier gets prompt caching on by default, per-key hard caps, signed receipts and zero prompt retention. A cheaper plan is a smaller allowance, not a worse product.
Work it out from the harness, not the headcount. A coding agent re-sends its system prompt and tool definitions every turn, roughly 25,000 tokens for Claude Code, so a working day of agent sessions is the unit that matters. Lite covers a few hours a day on a small repo; Pro covers a normal working day; Max covers all-day sessions across several repos.
If you are not sure, start on pay-as-you-go and read your own usage after a week. The dashboard shows spend per key and per model, which answers this better than any table here.
Plans do not replace prepaid credit. Buy credit when you want it, spend it at the same rates, and keep the same caps. The plan exists because a monthly number is easier to approve than an unpredictable one, not because the metering changes. See pricing for the per-token detail and spend caps for how a cap behaves at the edge.