Pay per token. Nothing else.
Prepaid credit, spent token by token. No subscription, no minimum spend, no invoice afterwards. Both model names, deepseek-v4.1-flash and deepseek-v4.1-flash-nothink, cost the same. These are launch prices: they may change after our benchmarks, and we email you at least 14 days before any increase.
| Per 1M tokens | Price | What it is |
|---|---|---|
| Input | $0.22 | Prompt tokens the cache hasn’t seen |
| Cached input | $0.045 | Repeated prefixes: system prompts, long chats, the same code sent again. About 80% off, automatically |
| Output | $0.75 | What the model writes, reasoning included |
- Prepaid in US dollars. Add credit from $5 to $5,000, paid in Bitcoin. Credit never expires.
- Stops at $0. New requests are refused when credit runs out; nothing is billed afterwards.
- A key for each app, each with its own spend limit if you like, and its spend shown separately.
Room for agents that run in parallel
How many requests an account can run at once, across all its keys, grows with the credit it has bought. It rises on its own when you top up.
| Level | Credit bought, all time | Requests at once | Per minute |
|---|---|---|---|
| Level 1 | Under $20 | 5 | 100 |
| Level 2 | $20 to $100 | 15 | 300 |
| Level 3 | $100 to $1,000 | 30 | 600 |
| Level 4 | $1,000 and up | 60 | 1,200 |
A request past the limit gets HTTP 429 with code concurrency_limit_exceeded and a Retry-After header; the OpenAI and Anthropic SDKs retry it by themselves. Your level and limit are on the Billing page.
Billing questions
Setup and technical questions are in the docs.
How does caching work?
It’s automatic. When a request starts with the same prefix as an earlier one, such as a system prompt or a long conversation, those tokens are billed at the cached price, 80% below the input price. Each account’s cache is isolated, so nobody else’s requests can hit yours.
What happens when my credit runs out?
New requests stop at $0. A request already running can finish, so a balance can dip a fraction of a cent below zero; your next top-up covers it first. There’s no bill afterwards, and credit never expires.
How do I pay?
Credit is held in US dollars and paid in Bitcoin. There’s no card on file and nothing recurring.
Can I get a refund?
Unused credit is refunded if you ask within 14 days of paying, and in a few other cases, such as a payment made by mistake. The refund policy has the details.
What if I need more requests at once?
Your limit rises with the credit you’ve bought (the table above). A request past it gets a 429 that SDKs retry automatically once one of your requests finishes.
Swap the base URL and keep building
Prepaid credit that never expires.