API vs subscription: which is cheaper?
Coding agent tools bill in two ways. A subscription charges a flat price per seat and limits how much you can use. The API charges for every token the agent sends and gets back. Which is cheaper depends on how much your agent works, and some “subscriptions” turn out to be API billing with a seat fee on top.
The two billing models
A flat plan has a known price and a usage allowance. Past the allowance you wait for it to reset or buy credits. Pay-per-token has no allowance and no ceiling: a busy month costs more, a quiet one costs less. Here is how each tool with a cost page bills:
| Tool | Seats | How usage is billed |
|---|---|---|
| Claude Code | $20 to $200 a seat | Flat plans with usage limits, an API key billed per token or a seat price plus every token at API rates |
| Cursor | $30 to $200 a seat | A seat price plus every token at API rates |
| GitHub Copilot | $10 to $100 a seat | Seats that include usage credit, then API rates |
When a subscription wins
A flat plan wins when the agent works most days and the plan's limits hold. At 2 medium tasks a day on Claude Opus 5.5, run carefully, API rates come to about $283 per developer a month. Claude Code Ultra costs $200, if that usage fits its allowance.
When the API is cheaper than a subscription
The API wins for light or occasional use. At API rates, the $20 Claude Code Pro price buys about 3.0 careful Claude Opus 5.5 tasks a month. If your agent does less than that, pay per token.
The break-even point
Medium agent tasks a month at which API rates cost the same as the plan, run carefully. Below the number, the API is cheaper. Above it, the plan is, until its limits.
| Plan | Seat | Claude Opus 5.5 | Claude Sonnet 5.5 | Claude Haiku 5.5 |
|---|---|---|---|---|
| Claude Code Pro | $20 | 3.0 | 3.2 | 2.3 |
| Claude Code Max | $100 | 15 | 16 | 11 |
| Claude Code Ultra | $200 | 30 | 32 | 23 |
| Claude Code Team, standard seat | $25 | 3.7 | 3.9 | 2.9 |
| Claude Code Team, premium seat | $125 | 19 | 20 | 14 |
GitHub Copilot Pro, Pro+, Max, Business and Enterprise include usage credit and bill API rates past it, so there is no crossing point: below the credit you pay for usage you didn't use, above it you pay the API price.
Claude Code Enterprise (billed annually, 20-seat minimum); Cursor Premium, Super and Ultra charge API rates for usage on top of the seat price, so the seat never lowers the token bill.
Claude Code, Cursor and GitHub Copilot side by side
Claude Code: subscription vs API
| Plan | Per seat a month | Beyond the plan |
|---|---|---|
| API key (pay as you go)Every token billed at API list price. | $0 | Every token at API rates |
| ProClaude Code within the Pro usage allowance; usage credits at API rates after that. | $20 | Usage limits, then optional credits at API rates |
| Max5× Pro usage every 5 hours and every week. | $100 | Usage limits, then optional credits at API rates |
| Ultra10× Pro usage every week, no 5-hour limit. | $200 | Usage limits, then optional credits at API rates |
| Team, standard seatClaude Code included; usage credits at API rates past the seat allowance. | $25 | Usage limits, then optional credits at API rates |
| Team, premium seat5× the usage of a standard seat. | $125 | Usage limits, then optional credits at API rates |
| Enterprise (billed annually, 20-seat minimum)Seat price plus usage at API rates. | $20 | All usage at API rates, on top of the seat |
Cursor: API vs subscription
| Plan | Per seat a month | Beyond the plan |
|---|---|---|
| PremiumWeekly pool 4× Plus for Grok/Composer; Claude usage at API list price. | $30 | All usage at API rates, on top of the seat |
| SuperWeekly pool 15× Plus for Grok/Composer; Claude usage at API list price. | $100 | All usage at API rates, on top of the seat |
| UltraWeekly pool 50× Plus for Grok/Composer; Claude usage at API list price. | $200 | All usage at API rates, on top of the seat |
GitHub Copilot: API vs subscription
| Plan | Per seat a month | Beyond the plan |
|---|---|---|
| Pro1,000 base AI credits ($10) a month, plus variable flex credits. | $10 | $10 of usage included, then API rates |
| Pro+3,900 base AI credits ($39) a month, plus variable flex credits. | $39 | $39 of usage included, then API rates |
| Max10,000 base AI credits ($100) a month, plus variable flex credits. | $100 | $100 of usage included, then API rates |
| Business1,900 AI credits ($19) per user a month, pooled across the organization. | $19 | $19 of usage included, then API rates |
| Enterprise3,900 AI credits ($39) per user a month, pooled across the organization. | $39 | $39 of usage included, then API rates |
List prices on monthly billing unless the plan says otherwise, checked 2026-10-31.
Run your own numbers
Estimate your month
Quote: 10 developers, 21 working days
- API key (pay as you go)$2,828$283 per developer
Every token billed at API list price.
Includes $283 per developer of token usage at API rates.
- Pro$200$20 per developer
Claude Code within the Pro usage allowance; usage credits at API rates after that.
At this usage ($283 at API rates) you would likely hit the plan's limits, then pay extra only if you turn on usage credits.
- Max$1,000$100 per developer
5× Pro usage every 5 hours and every week.
At this usage ($283 at API rates) you would likely hit the plan's limits, then pay extra only if you turn on usage credits.
- Ultra$2,000$200 per developer
10× Pro usage every week, no 5-hour limit.
At this usage ($283 at API rates) you would likely hit the plan's limits, then pay extra only if you turn on usage credits.
- Team, standard seat$250$25 per developer
Claude Code included; usage credits at API rates past the seat allowance.
At this usage ($283 at API rates) you would likely hit the plan's limits, then pay extra only if you turn on usage credits.
- Team, premium seat$1,250$125 per developer
5× the usage of a standard seat.
At this usage ($283 at API rates) you would likely hit the plan's limits, then pay extra only if you turn on usage credits.
- Enterprise (billed annually, 20-seat minimum)$3,028$303 per developer
Seat price plus usage at API rates.
Includes $283 per developer of token usage at API rates.
How this works
Seat prices are each vendor's list prices. Task costs are the average of 20,000 simulated runs of an agent on a 5-point ticket, priced at API list prices. Subscription allowances aren't published in dollars, so where a plan has limits we can't say exactly when you would hit them. Treat it as an order of magnitude. Method and sources.
This page is independent and not affiliated with any model or tool vendor.
See where the money goes, one run at a time.
Play the $50 agent game