Track · Product Strategy8 min read
How to choose an AI plan for a team: five questions
Every price, quota and rule below was read on the vendor's own page on 19 or 20 November 2026. Three links are referral links: if you use them we may earn a commission or credits, at no extra cost to you. Every other link goes to the vendor and earns us nothing.

A coding plan is a capacity decision, not a discount: you buy a ceiling on how much work the team can push through a model in a window. Buy seats before the load is measured and you have prepaid an appetite you cannot see.
The second decision is ownership. A seat plan belongs to a named person who runs a tool in the loop; a shared pool belongs to the organisation, with one invoice and one set of rules on concurrent requests.
What does the team actually use in a five-hour window?
Publish three numbers before a price: tokens per five hours, tokens per week, and the cache hit rate behind both.
Z.ai publishes two credit windows per tier: Lite 2,000 credits per five hours and 10,000 a week, Pro 12,000 and 60,000, Max 28,000 and 140,000. MiniMax counts the same way — a five-hour and a seven-day window that reset independently, with unused quota carrying over to neither window nor billing period. Kimi publishes the opposite shape: a rolling five-hour limit, no weekly limit on current plans, one shared monthly pool.
How much of that traffic is cache reads?
Cache reads are the cheapest lever you control, and vendors price them separately. Z.ai publishes the credit formula for GLM-5.3: input tokens count at 6.9, cached input at 1.7, output at 24, divided by 10,000 — a cached token at about a quarter of an uncached one (our arithmetic). Off-peak usage is charged at half the standard rate; peak hours run Monday to Friday, 14:00-18:00 Singapore time, so weekends are entirely off-peak.
Ollama's price list is per million tokens: GLM-5.3 at $1.40 input, $0.26 cached, $4.40 output. Worked example (our arithmetic): 10 million input tokens a month, 80 % of them cache reads, cost $4.88 against $14.00 uncached, output excluded.
Decide with your measured cache hit rate and the published cached-input price.
Who is allowed to run the plan?
MiniMax says its plan is for individual, interactive developer use and points production at pay-as-you-go. Z.ai restricts the plan to officially supported tools and products: the subscription is tied to one natural person with no account sharing, and only two models can be called.
Alibaba Cloud's Token Plan is for interactive use only in coding tools and agent tools, and must not be used for automation scripts or non-interactive batch calls; the page says that may mean suspension or an API key ban. Openference's terms cover personal, interactive, human-in-the-loop sessions and exclude background agent loops, batch pipelines and headless daemons.
A job somebody waits for is a seat; a job that runs while nobody watches is metered per token.
What does the plan not cover?
- Credits are not app limits. OpenAI's API credits do not cover Codex through API keys and do not raise Codex limits. Anthropic's credits cover the API, Managed Agents, the Agent SDK and the playground, not Claude Code, and do not change your usage limits in Claude.
- The model list is shorter than the catalogue. Z.ai's plans carry GLM-5.3 and GLM-5.3-Flash only; older models reroute to them.
- Nothing rolls over. Anthropic's credits expire at the cycle end, Ollama's included usage refreshes monthly, MiniMax's unused quota carries to neither window nor period, and Kimi's pool resets with no rollover.
- Seats are not pools. Anthropic pools Team credits under a cap, so extra seats do not lift the ceiling; Ollama sells unlimited users on one shared balance.
What do you do with the volume you waste?
MiniMax sells Credits separately — $5 for 5,000 credits, $25 for 25,000, $100 for 100,000, valid for a year — and draws them automatically at the plan limit. Kimi's Extra Usage wallet takes a minimum 25 yuan top-up, caps the balance at 10,000 yuan, never expires and is deducted last.
Z.ai offers no paid overage: when the quota is gone you wait for the next five-hour cycle and the system does not touch your balance. The only extra volume is free campaign levers — quota reset cards and referral credits the vendor says never expire.
So: measure the load, fix the waste, then decide whether the ceiling is the problem.
What each profile should buy
| Team profile | What to buy | Published monthly price | Published numbers the decision rests on |
|---|---|---|---|
| One developer in the IDE | GitHub Copilot Pro | $10 per user per month | $15 of monthly credits; unlimited code completion and next edit suggestions |
| Five people, agents at the desk | Kimi Max is a personal plan with no seats, so pay-as-you-go for the rest | Kimi Max $99 a month, one account; Z.ai Pro $80; MiniMax Explore $55 | Kimi Max gives 5x agent credits and up to 4 parallel agent tasks, a rolling five-hour limit, no weekly limit, a shared pool and no rollover |
| Thirty people across teams | Ollama Team or Z.ai team seats | Ollama Team $500 a month; Z.ai Standard seat $88, Premium $188 | Unlimited users and $1,000 of usage credits shared across the team, centralised billing, 10 concurrent requests; Z.ai Standard seat 66,000 credits a week, Premium 155,000 |
| Agents around the clock | No subscription: a metered key | Metered per token at the vendor's list price | MiniMax points production at pay-as-you-go; Z.ai needs a written agreement for your own apps; Openference rejects unattended automation |
When the right answer is to buy nothing
- When the load is small and lumpy: a subscription buys a floor you may never use, and unused quota expires.
- When the work is unattended or unmeasured: the terms do not cover it, and a week of measurement beats a year of prepaid seats.
Frequently asked questions
Do I need a seat-based plan or a shared pool? Seats for named people, a pool for the organisation. Z.ai ties a subscription to one natural person and forbids account sharing; Ollama runs one account per person. Ollama's team plan carries unlimited users on one shared balance, Anthropic pools team credits under a cap, Kimi draws every feature on one pool.
What happens when a team hits the quota mid-launch? It depends on the vendor, so it belongs in the request. Z.ai makes you wait for the next five-hour cycle and does not bill your balance; MiniMax draws on purchased credits first, then pay-as-you-go; Kimi waits for the monthly refresh or uses Extra Usage; Ollama draws on a shared balance you can cap.
Do unused credits roll over to the next month? Mostly no. Anthropic's credits expire at the end of each billing cycle, Ollama's included usage refreshes monthly with no rollover, MiniMax's unused quota carries to neither window nor period, and Kimi's monthly pool resets with nothing carried over. Two exceptions: MiniMax's purchased credits last a year, Kimi's Extra Usage balance never expires.
How do I choose a tier without measuring anything? You cannot, and the vendors make that harder. Z.ai publishes absolute credit counts per window, so a tier can be sized from your own usage. MiniMax publishes only multipliers against its entry tier; Kimi publishes neither a numeric five-hour allowance nor a monthly total — those live in each console.
Is the annual plan worth it? Only after a measured month. Z.ai's yearly billing carries 30 % off, Kimi advertises up to $480 a year saved, and MiniMax's annual price pays for ten months and gives two free. All three turn a monthly option into twelve months of commitment on an unmeasured load profile.
Sources
Vendor pages, read 19 or 20 November 2026 (Openference on 20 November): Z.ai overview · Z.ai FAQ · MiniMax usage rules · MiniMax FAQ · Kimi pricing · Ollama pricing · Copilot plans · OpenAI API credits · Anthropic API credits · Alibaba Cloud Token Plan · Openference terms
Referral links
- CometAPI — https://www.cometapi.com/?ref=evgeny&utm_source=media&utm_medium=article&utm_campaign=ai-plan-for-team-roadmapfile&utm_content=ref-cometapi
- NanoGPT — https://nano-gpt.com/r/yUf6yXwN?utm_source=media&utm_medium=article&utm_campaign=ai-plan-for-team-roadmapfile&utm_content=ref-nanogpt
- Z.ai — https://z.ai/subscribe?ic=FVQVPVL9L0&utm_source=media&utm_medium=article&utm_campaign=ai-plan-for-team-roadmapfile&utm_content=ref-zai
CometAPI pays 20 % of every referred top-up and gives the reader nothing; NanoGPT pays 10 % of eligible usage plus 5 % of subscription purchases and gives the reader 5 % off eligible usage; Z.ai pays credits worth 10 % of the first payment and takes 10 % off the first subscription, for a new user who pays within 72 hours.
Affiliate disclosure: the three links above are referral links. A qualifying CometAPI top-up may earn us a commission and eligible NanoGPT usage may earn us credits, at no extra cost to you; the Z.ai link takes 10 % off your first GLM Coding Plan subscription and pays us credits worth 10 %. Every other link goes to the vendor and earns us nothing. Prices were correct on 19 and 20 November 2026.
More from Priya Raman
Show full bio
News editor covering money, funding and market shifts around product management at Roadmap File.
78 articles