Parallel agents quota and cost: one pool, N contexts

Do not expect that; the official text makes no such promise. What the docs specify is that Claude Code blocks further requests until the reset time shown in the message, and swapping models does not help either: session and weekly limits are shared across all models. The sanctioned way to resume is to leave the session open on a subscription and continue the interrupted task shortly after the reset — the same subscription waits out the reset, it does not hop accounts. More importantly the official usage policy lists Create or manage multiple accounts to evade detection or circumvent platform safeguards as prohibited, so switching account at the cap is not an official route. The parallel side is the same: a few more agents do not open a second line, they hit the same 5-hour window and the same weekly limit. The official text describes subscription quota through how many messages you can send and how long you can work, and gives no formula in tokens, so any conversion is checked against the status bar and the provider's bill. The weekly layer, per the docs, resets at a fixed time each week that is assigned to your account, independent of when you started. Checked on 2026-10-20.

For models the wording is explicit: each model has its own prompt cache, so the next request re-reads the whole conversation with no cache hits, even when the content is word for word identical. For accounts, what the docs actually say is that caches are isolated between organizations and that different organizations never share caches, even if they use identical prompts — move to another organization and the same prefix is written again, at 5-minute cache write tokens are 1.25 times the base input tokens price. Even staying put requires the prefix to match 100% identical prompt segments, so a branch name or a per-worktree path inside the prefix defeats the hit. For worktrees the official wording is that each worktree is a separate working directory with its own files and branch. The community talk of a bill exploding once the quota is burned has no official counterpart in the official pages: it is community feedback without an official wording. Checked on 2026-10-20.

Not necessarily; often the opposite. Every agent carries its own copy of the fixed prefix: each subagent starts with a fresh, isolated context window, and each teammate has its own context window and consumes tokens independently, with token usage roughly proportional to team size; the workflows page adds that a single run can use meaningfully more tokens than working through the same task in conversation. Under the assumption P = 40,000, splitting one job into K parts pays that prefix K times — our arithmetic, not an official figure. The only multiplier Anthropic itself gives is plan mode agent teams at 7x more tokens than standard sessions, and that is tokens, not a bill seven times larger. The break-even test is one line: whether the duplicate reads you avoid exceed the extra prefixes you pay for. Checked on 2026-10-20.

Start with provenance: the official text warns that a window is almost used up and sets no percentage of its own — 80 percent is not official wording, and the official pages give no 80% threshold — while Anthropic's own example wording is 85% of your session limit. The layer it watches is a pair, five-hour session and weekly usage limits, defined mechanically as the percentage of the 5-hour or 7-day rate limit consumed, from 0 to 100, and those windows appear only for claude.ai Pro and Max subscribers, so an API key has neither. So read the number only after checking which credential you are on. Fine as a brake light, wrong as a reconciliation basis: the official documentation has no fixed percentage threshold, only this kind of warning. Checked on 2026-10-20.

Two clocks again. On the API side limits are measured per model class in RPM, ITPM and OTPM, and going past them returns 429 with a retry-after; two official sentences land straight on parallelism: Short bursts of requests can exceed the limit and trigger rate limit errors, and a sharp jump in usage can also hit acceleration limits, which the official text answers by ramping traffic up gradually. That pool is sized where Limits are set at the organization level, so a fan-out never enlarges it, and capacity is refilled continuously rather than being reset at fixed intervals. On the subscription side the cap is not a 429: Claude Code blocks further requests until the reset time shown in the message. For API keys, cloud hosting and pay-as-you-go the docs say usage there is metered per request, so there is no reset to wait for — there is nothing to wait out. Our own key allows concurrency 5; the extra agents queue, and queueing is not a ban. Checked on 2026-10-20.

Follow the three-step example above, then add three limits. First, dates on unit prices: the official pricing page itself states that prices and plans are subject to change at Anthropic's discretion, and the cache tier carries the 5-minute TTL note. Second, verify hits yourself: cache hits require 100% identical prompt segments, so one differing byte in the prefix means no hit. Third, reconcile against the provider's bill or against this site's key: claude.ai, Claude Code and Desktop share one allowance on an account, so you assemble the provider or key bill, not a sum of status bars. The token volume, the agent count and the turn count in the example are assumptions set for it, so do not read them as official numbers when you copy the example; prices follow the price list, limits follow the subscription, and the two must not be mixed. Checked on 2026-10-20.