Per-Seat vs Token-Based AI Pricing: Which Is Cheaper for 1,000 Employees?

In per-seat vs token-based AI pricing for 1,000 employees, per-seat licenses are cheaper when most people use AI daily, and token-based pricing is cheaper when use is light or concentrated in a few teams. At list prices of $20 to $30 per user per month, 1,000 seats cost $240,000 to $360,000 a year before usage and add-ons, so the break-even depends on how many seats stay active and which models people use.

The arithmetic: $20 x 1,000 seats x 12 months = $240,000 (Claude Enterprise, before usage), and $30 x 1,000 x 12 = $360,000 (Microsoft 365 Copilot, before the Microsoft 365 plan it requires). I read every vendor price on this page from the vendor's own site on October 29, 2026.

Coworker is a context layer for enterprise AI: it connects 50+ work tools, builds a permission-aware company graph (OM2) and runs agents that act across them, inside Coworker or Claude, ChatGPT and Cursor via Coworker MCP. This page compares the two pricing models, not vendors. For vendor detail, see my Claude Enterprise pricing breakdown, the enterprise AI pricing comparison and the quarterly enterprise AI price index.

How do per-seat vs token-based AI pricing models differ?

A per-seat license charges a fixed fee per named user per month, whether that person sends one message or a thousand, usually with usage limits on top. Token-based pricing charges for the text a model reads (input tokens) and writes (output tokens), usually quoted per million tokens, so the bill tracks actual use.

Most enterprise plans now mix the two. Claude Enterprise charges a seat fee and bills all usage at API rates on top. Microsoft offers Copilot Chat at no additional cost with eligible Microsoft 365 plans, sells Microsoft 365 Copilot as a $30 add-on, and meters agents separately. OpenAI's Enterprise plan is priced by quote, and its pricing page says credit-based and token-based pricing are both available. My price index tracks this shift: half the platforms in it carry a consumption meter somewhere in the bill.

That last row matters. A token price buys model output, not a product. To give 1,000 people token-priced AI you also need an interface, connectors to company data, access controls and spend limits.

How much do 1,000 AI seats cost at published prices?

Between $240,000 and $360,000 a year at the standard published seat prices, before usage, prerequisite licenses and add-ons. Two common plans do not take 1,000 people at all, so at that size you are choosing between enterprise plans, several of them quote-only.

Microsoft 365 Copilot's $30 is the AI increment, not the total, because the required Microsoft 365 plan is billed separately. Claude Enterprise's $20 seat buys access only, and usage scales with model and task. My Claude Enterprise pricing breakdown models that usage for 50, 300 and 1,000 people, so I do not repeat it here.

What does the same work cost on tokens?

On token pricing, 1,000 light users cost between $2,400 and $96,000 a year, depending only on which model they use. I use the same usage profile as my AI token cost comparison: 1 million input tokens and 200,000 output tokens per person per month, and a heavier profile of 10 million input and 2 million output.

List prices, per million tokens, from each provider's own page on October 29, 2026: Claude Haiku 5.5 $0.10 input and $0.50 output (prompts up to 100,000 tokens), Claude Sonnet 5.5 $2 and $10, Claude Opus 5.5 $4 and $20, and GLM 5.3 Flash $0.15 and $0.50.

Each yearly figure is the monthly figure x 1,000 users x 12 months. These are list prices with no prompt caching or batch discounts, and they cover tokens only, not the interface, connectors or admin tooling a seat includes. Use the LLM cost calculator to rerun them on your own volumes.

Two things stand out. Light users are far cheaper on tokens than on any seat, even on Claude Opus 5.5 ($8 a month against a $20 or $30 seat). Heavy users on a frontier model cost more on tokens than a seat: $40 a month on Sonnet 5.5 and $80 on Opus 5.5, against $30. That is why the answer turns on how usage is spread across your 1,000 people.

How many tasks a month does a seat have to replace to break even?

Divide the seat price by the cost of one task. Artificial Analysis, the independent benchmark site, measures cost per task as the average cost to run one task from its Intelligence Index evaluation set at each provider's prices. My AI token cost comparison lists those figures for 13 models, read October 15, 2026: $0.21 per task for Claude Haiku 5.5, $0.28 for GLM 5.3 Flash, $0.72 for GPT 6.1 Sol, $5.46 for Claude Sonnet 5.5, $5.98 for Claude Opus 5.5 and $7.63 for Claude Fable 5.1.

These are Artificial Analysis benchmark tasks, not your company's tasks. They run each model at its highest reasoning setting on hard evaluation problems, so a routine lookup or summary costs far less. What the table shows reliably is the ratio: the same $30 buys roughly 29 times as many tasks on Haiku 5.5 as on Opus 5.5 ($5.98 / $0.21 = about 28.5).

Another way to see the break-even is in tokens. At a 5-to-1 input-to-output mix, $30 buys 3.75 million input plus 0.75 million output tokens on Claude Opus 5.5 ((3.75 x $4) + (0.75 x $20) = $15 + $15 = $30), and 150 million input plus 30 million output on Claude Haiku 5.5 ((150 x $0.10) + (30 x $0.50) = $15 + $15 = $30). A person who uses less than that in a month is cheaper on tokens than on a $30 seat.

What share of seats sits idle, and how do you find out?

I could not find a published, AI-only figure for idle seats. The closest dated number is Zylo's 2026 SaaS Management Index, released January 29, 2026 and built on more than 40 million SaaS licenses: measured against industry-recommended utilization levels, organizations leave an average of 36% of their SaaS licenses unused. That covers all SaaS, not AI assistants specifically.

If 36% of 1,000 seats went unused, the cost per active seat rises sharply:

• Active seats: 1,000 x (1 - 0.36) = 640
• At $30 per seat: $360,000 / 640 = $562.50 a year, or $46.88 a month per active user
• At $20 per seat: $240,000 / 640 = $375 a year, or $31.25 a month per active user

To get your own number, use the vendor's admin reports rather than a survey. Microsoft's Copilot usage report shows enabled users, active users and the active-user rate, and OpenAI's pricing page lists usage analytics on ChatGPT Business and user analytics on Enterprise. Pull 90 days, count people below a usage threshold you choose, and move them off seats. Every seat price in the table above is an annual price, so plan the cut for renewal.

When does a hybrid of seats and usage win?

A hybrid wins when usage is uneven, which it usually is in a 1,000-person rollout: give seats to the people who use AI heavily every day and put everyone else on usage. Here is an illustration with made-up proportions, not a measurement: 200 heavy users and 800 light users, using the per-person costs from the token table above.

In this example the hybrid is cheapest, because each heavy user costs $30 on a seat instead of $40 on Sonnet 5.5 tokens, and each light user costs $4 on tokens instead of $30 on a seat. The tokens-only rows leave out the interface and admin tooling a seat includes, and the seat row assumes nobody hits a usage limit. Change the mix and the order changes: if most of your 1,000 people are heavy daily users, all-seats moves ahead.

What lowers both bills?

Four levers apply whichever way you buy:

1. Right-size seats at renewal. Use 90 days of admin data to drop or downgrade seats below your usage threshold.
2. Make a low-cost model the default. On the Artificial Analysis figures above, Opus 5.5 costs about 21 times as much per task as GLM 5.3 Flash ($5.98 / $0.28 = about 21.4). Most lookups and summaries do not need the top model. My model routing page shows how Coworker sets a default and lets people switch.
3. Use caching and batch pricing on metered lines. Anthropic's pricing page lists cache reads well below input rates and 50% off with batch processing.
4. Send each request less context. Every token a model reads is billed on a metered line. My guide to AI cost optimization covers the full list of levers.

Set spend limits on any usage line before rollout. Claude Enterprise admins can set user and organization spend limits, per Anthropic's pricing page.

Where does Coworker fit in a per-seat or token-based rollout?

Coworker works on the token side of the bill: which model reads each request, and how much company context it has to read. It does not change a vendor's seat price.

Every new chat in Coworker starts on Optimized, which runs GLM 5.3 Flash today. Anyone can switch a chat to a frontier model such as Claude Opus 5.5, and admins set the workspace default and can block models they do not want in use. OM2 narrows each task to the records that matter across 50+ connected tools before a model sees it. On Coworker's published 100-task benchmark against Claude with its native connectors, that cut token cost 9x on retrieval-heavy tasks with OM2 context alone, with no change to the model, and 51x with OM2 plus model routing. Blind reviewers preferred the OM2 answers in 84.5% of pairwise comparisons. Coworker ran that benchmark itself, so weigh it as a vendor result.

If you already pay for Claude or ChatGPT seats, Coworker MCP brings the same company context into those assistants. To test the effect on your own data, Coworker's Context Impact Report runs the benchmark on your systems; my guide to benchmarking enterprise AI platforms explains the method.

Coworker is enterprise-sales-led. Its pricing page scopes each plan to the connectors, people and deployment you need, with no public price list. To see what it would cost for your company, book a demo.

Frequently asked questions

Per-seat AI licenses or token-based pricing: which is cheaper for 1,000 employees?

Token-based pricing is cheaper when most employees are light users, and per-seat licenses are cheaper when most use AI heavily every day. At 1 million input and 200,000 output tokens per person a month, 1,000 people cost $4,000 a month on Claude Sonnet 5.5 list prices, against $30,000 a month for 1,000 seats at $30.

How much does it cost to give 1,000 employees AI seats?

At published list prices, $240,000 a year for Claude Enterprise seats at $20 per month before usage, and $360,000 a year for Microsoft 365 Copilot at $30 per month before the required Microsoft 365 plan. ChatGPT Business caps at 200 people and Claude Team at 150, so larger companies get enterprise quotes.

What is the break-even between a seat and token pricing?

Divide the seat price by your cost per task or per token. A $30 seat equals about 143 Artificial Analysis benchmark tasks on Claude Haiku 5.5 at $0.21 each, or about 5 on Claude Opus 5.5 at $5.98. In tokens, $30 buys 3.75 million input plus 0.75 million output on Opus 5.5.

How many AI seats go unused?

There is no published AI-only figure that I could find. Zylo's 2026 SaaS Management Index, released January 29, 2026, found organizations leave an average of 36% of SaaS licenses unused against recommended utilization levels. Check your own vendor's admin usage report before renewal.

Is a hybrid of seats and usage cheaper?

Usually, when usage is uneven. In an illustrative 1,000-person split of 200 heavy and 800 light users, seats for the heavy users plus Sonnet 5.5 tokens for the rest came to $110,400 a year, against $360,000 for 1,000 seats at $30. Your own usage mix decides the result.

Does Coworker charge per seat or per token?

Coworker has no public price list. Pricing is scoped to the connectors, people and deployment you need, so book a demo for a quote. On cost, Coworker's role is the token side: an Optimized default model, admin model controls and OM2, which cut token cost 9x on retrieval-heavy tasks in Coworker's benchmark.

Related reading

• Claude Enterprise pricing: seats, usage and costs for 50 to 1,000 people
• Enterprise AI pricing compared
• Enterprise AI price index, Q3 2026
• AI token cost: 13 models compared by cost per task
• How to benchmark enterprise AI platforms before a POC
• AI cost optimization: where the money goes and 9 levers that cut it
• How much should enterprise AI cost per user?
• Microsoft 365 Copilot enterprise pricing
• ChatGPT Enterprise pricing
• Gemini Enterprise pricing
• LLM cost calculator
• Model routing
• Coworker benchmarks