AI bills are shaped by what you pay for—user access, measured consumption, a recurring subscription, or some combination—not just by the plan’s headline price. A per-seat fee may cover access but leave token use extra; usage-based billing rises or falls with activity; and a flat monthly price may still come with limits. Compare the billable unit, included usage, and overage rules before estimating what a plan will cost your budget.
What the main AI pricing models mean
These labels describe billing mechanics, not mutually exclusive plan types. A single service can charge for seats and then meter usage, or combine a subscription with credits and limits. The important question is what triggers each charge.
Per-seat pricing
A per-seat plan charges for each licensed user over a billing period. It can make the access portion of a bill easier to forecast when team size is stable, but it does not necessarily include each person’s AI consumption. For example, Anthropic’s Enterprise help page says seats provide access and token use is billed separately at standard API rates. Anthropic’s Enterprise usage documentation describes usage charges based on actual token consumption.
Usage-based pricing
Usage-based billing charges for a measured unit. Depending on the product, that unit may be input, cached-input, or output tokens; a fixed number of credits per message or task; a generation; or connected minutes. The bill depends on the amount and kind of activity, the model or feature used, and the applicable rates. OpenAI’s business and Enterprise/Edu credit rate card illustrates both fixed-credit charges for some activities and token-based credit charges for others. The rate card that applies is determined by the customer agreement. OpenAI’s credit documentation describes these units.
#1 Best Overall
Flat-rate or subscription pricing
A recurring subscription price can simplify the base budget, but “flat-rate” does not by itself mean unlimited use. Plans may impose session windows or other caps, stop or restrict work at a limit, or let customers buy additional usage. Claude’s plan documentation describes rolling session windows, additional caps, and optional usage credits after limits are reached. Check the plan’s current terms rather than assuming the subscription covers every workload. Claude pricing and plan details show the published structure.
Hybrid pricing
Hybrid plans combine these mechanics: for example, a seat or subscription charge plus metered tokens, prepaid credits, usage caps, or discounts tied to a spending commitment. Treat “per-seat,” “usage-based,” and “flat-rate” as clues about the bill, not complete descriptions of it.
How the bill can change with workload
Two teams with the same number of seats can have very different bills if one sends more requests, uses different models, or generates longer outputs. Likewise, two workloads with similar request counts can cost differently when their input and output sizes or model choices differ.
OpenAI’s eligible Enterprise token-based rate card lists prices in U.S. dollars per million tokens. At the time the page was inspected for this article on October 7, 2026, it listed:
| Model on the rate card | Input per million tokens | Cached input per million tokens | Output per million tokens |
|---|---|---|---|
| GPT-6 Astra | $10 | $1 | $50 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
These are volatile, plan-eligible rate-card examples, not a market-wide comparison or a forecast for a particular customer. The page says actual costs vary with model, task size, input/output mix, automations, fast mode, and concurrent instances. The applicable rate card depends on the customer’s plan or agreement. Check the live page and contract before using any figure in a budget. OpenAI’s Enterprise and Edu model rate card provides the listed rates and qualifications.
Compare plans by the costs and limits that matter
Before comparing monthly totals, identify every charge and what happens when included usage runs out.
Rank #4
- Billable unit: Is the charge per named user, token type, request, minute, credit, or committed spend? Check whether input, cached input, output, tools, and agent activity have different rates.
- Included allowance and limit behavior: What usage is included? Do limits reset, pool across users, or apply separately? At the limit, does work stop, incur overages, or require purchased credits?
- Seat versus consumption: Does a seat include usage or only platform access? Anthropic’s current Enterprise description separates its seat fee from token consumption.
- Workload sensitivity: Estimate using representative input and output sizes, model mix, caching, reasoning or fast modes, automations, and concurrency. A per-token rate alone does not predict the team’s bill.
- Budget controls and timing: Look for user- or organization-level spending caps, usage visibility, and whether credits are prepaid or usage is billed afterward. Anthropic says self-serve Enterprise usage is purchased upfront in shared credits, while sales-assisted usage is billed monthly in arrears. Anthropic’s Enterprise usage documentation sets out those billing arrangements.
- Commitment and eligibility: For a discount tied to committed spend, confirm the term, eligible products or SKUs, spend window, exceptions, and cancellation rules.
Estimate a realistic monthly cost
Use your own workload rather than multiplying a headline price by a guessed amount of “average” usage. Build light, typical, and heavy scenarios using expected users, model choices, input and output volume, and likely concurrency. Apply the relevant rates and include subscription or seat charges, credits, and any committed spend. Then compare what happens in each scenario when usage reaches plan limits.
For example, a stable headcount can make seat costs predictable while a surge in token use still raises the variable portion of the bill. A subscription with caps can keep the base charge steady but constrain work or lead to paid credits when activity is high. These scenarios are planning methods, not guarantees of a provider’s final charges; negotiated terms and the customer’s actual usage matter.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →When a committed-spend discount is worth considering
A commitment can reduce eligible usage rates, but it trades flexibility for a term-bound spending obligation. Google Cloud says its Flexible Savings Plans require a specific monthly spend commitment over a one- or three-year term. Its documentation states that eligible Gemini Enterprise SKUs receive a 10% discount with a one-year plan or 20% with a three-year plan, subject to exceptions; the commitments cannot be cancelled, and third-party products do not receive the discount. These are Google Cloud terms, not general AI-market discounts. Verify eligible SKUs and final pricing before committing. Google Cloud’s Flexible Savings Plans documentation explains eligibility and commitment rules.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
