API billing
Pay-per-token billing on the Developer Platform — entirely separate from Claude app subscriptions, with its own limits, its own console and its own bill.
Last verified August 21, 2026 · 1 source
The most common billing misunderstanding in this ecosystem: a Claude subscription is not API credit. Pro, Max, Team and Enterprise govern the Claude apps. The API is billed separately, per token, with its own rate limits and its own console.
Having both is normal. Having one does not discount the other.
How it works
- Billed monthly on actual usage, in USD
- New accounts receive a small amount of free credit for testing
- Credit card for standard accounts; invoicing for enterprise
- Usage tracked in the Console
The discounts, and how they stack
| Lever | Effect |
|---|---|
| Prompt caching | Cache hits at 0.1x base input |
| Batch API | 50% off input and output |
| Model choice | Haiku 4.5 is a fifth of Opus 5 on input |
Caching and batching combine. On Haiku 4.5, batched cache hits land at $0.05 per million input tokens — a hundredth of Opus 5’s standard rate.
Two multipliers go the other way: inference_geo: "us" at 1.1x, and
fast mode at 2x.
Rate limits are not spend limits
Rate limits throttle throughput. They do not cap your bill. Spend controls are separate, available through the Console and on Enterprise.
Volume and academic pricing
Volume discounts are negotiated case by case; academic and research discounts may be available. Neither is published as a rate card.
Comparisons
Sources
- 01Pricing — Claude Platform DocsOfficialplatform.claude.com
Page changelog
- VerifiedBilling model and discount stacking re-checked.