Buying an AI coding API for a team:
quota, concurrency and sub-keys
Individual plans allow 5 concurrent requests per key; a team account allows 35 and scales linearly with account count, backed by a pooled $4,000 monthly quota with no daily cap. Budget roughly 3 people per account for a busy team, 5 for a normal one.
Four numbers to settle first
Pooled quota per account
Resets monthly with no daily cap; whatever is unused does not roll over.
Per-account concurrency
Scales linearly with account count; an individual plan gives 5 per key. Request rate itself is uncapped.
Suggested per account
About 3 per account for a busy team, 5 for a normal one. A starting estimate, not a hard limit.
Time to provision
Top up, email us, and it is handled by hand within about an hour during business hours.
For a team, the difference is shape, not price
An individual plan is prepaid against a daily cap: one ceiling per day, and whatever you do not use is gone. A team account is pooled monthly: one total for the month with no daily ceiling, so whoever works hardest simply draws more. For teams with uneven load, the second shape is far less likely to wall everyone off on the day one person is pushing. The second difference is concurrency: 5 per key on an individual plan versus 35 per team account, scaling with account count. Multi-agent orchestration, batch jobs and parallel CI usually hit that number long before they hit the quota.
Team plan terms in force (2026-09-02)
Start from a single account — there is no minimum purchase — and add more as the team grows. Each account carries a pooled $4,000 monthly quota and 35 concurrent requests. Volume tiers by account count: CNY 1,800 per account per month at 1–2, 1,760 at 3–6, 1,720 at 7–10 and 1,680 from 11 up, plus a one-time CNY 200 setup fee. Corporate bank transfer and VAT invoicing are supported, with the invoice fee banded by invoice value. One account draws on the same quota across Claude Code, Codex, Gemini CLI, OpenCode, Cline and Aider.
From estimate to provisioning
Estimate the account count: headcount divided by people per account (about 3 when busy, 5 otherwise), rounded up. If you are unsure, run one account for two weeks and add more against real consumption — accounts can be added at any time.
Work out the total: accounts × tier price × months, plus the one-time CNY 200 setup fee and the invoice fee if you need one. The calculator on the team plan page does this and shows how much your tier already saves against the 1–2 tier.
Pay and provision. Small amounts can be self-served; company payments go by bank transfer. Email us once funds land and it goes live in about an hour during business hours, with your own domain and sub-key management.
Confirmed terms vs common misreadings
Confirmed
① One account is enough to start, with no minimum. ② Each account has a pooled $4,000 monthly quota and 35 concurrent requests, scaling with account count. ③ Four volume tiers, down to CNY 1,680 from 11 accounts. ④ One account draws the same quota across tools. ⑤ Free replacement if an account is banned.
Common misreadings
① "Pooled means unlimited" — no, $4,000 per account per month is a ceiling and unused quota does not roll over. ② "Add quota to get more concurrency" — no, quota and concurrency are separate dimensions; concurrency comes from adding accounts. ③ "The team plan has a lower unit price than the monthly plans" — no. It is pooled monthly with no daily cap, so its unit rate sits slightly above the daily-cap plans; what you buy is pooling and concurrency. ④ "You must buy the enterprise plan to try it" — no, the cheapest individual plan has the same functionality.
Two ways to give a team quota
One individual plan per person
Everyone gets their own daily-cap plan. The accounting is as simple as it gets; the cost is that nobody's daily cap is fully used and none of it can be shared, and each key allows only 5 concurrent requests, which multi-agent work hits quickly. Fine for a small team with steady load.
Team accounts plus sub-keys
Buy pooled quota by account and issue each person a sub-key. Quota can be shared within the team, concurrency scales with account count, and consumption is visible per key; the cost is that you must estimate the account count up front and unused monthly quota does not roll over. Better for larger teams with uneven load.
How to attribute usage to people
Issue a separate sub-key per member — or per project, or per CI pipeline — instead of sharing one key across the group. That is what makes per-key consumption visible in the console and lets you say where the money went at month end. In practice, split along at least three axes: person (who is using it), environment (local development versus CI), and project (whether cost should be charged back to a business line). The finer the split, the faster an anomalous burn traces back to its source.
On QCode
The team plan page has a calculator: enter account count and duration and it returns the total, the monthly average, how much your tier already saves against the 1–2 tier, and how many more accounts reach the next tier — including what that costs per month and how much extra quota it buys. It also converts $4,000 a month into an approximate token count and a number of heavy developers it would sustain for a full month, so an abstract allowance maps onto a real team size.
FAQ
What is the minimum number of accounts?
One. There is no minimum purchase. Add accounts as the team grows; a new account is usable the same day.
How many people can share one account?
It depends on intensity. Budget about 3 per account for a busy team and 5 for a normal one; the hard figure is 35 concurrent requests per account, which usually serves 3–7 people. Treat it as a starting estimate and revise against two weeks of real consumption.
Does unused quota roll over?
No. The $4,000 per account per month is a pooled monthly allowance that resets; whatever is unused is not carried forward. So do not over-estimate the account count — start low and add.
Concurrency is not enough. Does more quota help?
No. Quota and concurrency are separate: quota decides how much you can run in total, concurrency decides how many run at once. Concurrency only comes from more accounts — 35 each, scaling linearly. The request rate itself is uncapped.
Can one account serve both Claude Code and Codex?
Yes. A single account draws on the same quota across Claude Code, OpenAI Codex, Gemini CLI, OpenCode, Cline, Aider and other mainstream tools — no separate purchase per tool.
Can we trial it small first?
Yes, and we suggest it. Buy the cheapest individual plan and validate the integration and stability first — functionality is the same, and what differs is the daily allowance and the team features (own domain, sub-key management, 35 concurrent). Decide on account count afterwards.
Sources
Quota, concurrency, volume tiers and provisioning time: the plan description and FAQ on /enterprise (checked 2026-09-02). Shared quota across tools and free replacement: the same FAQ. The 5-concurrent figure for individual plan keys: the comparison note on that page. The pages themselves govern at the time you read them. Seat and usage policies at Anthropic, OpenAI and Google are governed by their own help centres; this page does not restate their numbers.
Size the account count first
The team plan calculator returns the total, the monthly average, the tier discount and the token equivalent together.
Further reading
Team plans and the calculator
Volume tiers, pooled quota conversion and the provisioning flow, with a cost calculator.
AI API invoicing and corporate payment
The two VAT invoice fee bands, bank transfer and enterprise procurement.
Claude Code cost optimisation
How caching, context and model choice move the cost of a single call.
The quota, concurrency, prices and provisioning times on this page reflect the terms in force on 2026-09-02 and may change; the team plan page and our email confirmation govern. People per account is a planning estimate, not a performance commitment.