Home / Pricing
DeepSeek API pricing: V4.1-Flash 40% below DeepSeek's own price
Published 2026-10-16. DeepSeek's prices are taken from its pricing page (checked 2026-10-16). All prices are USD per 1M tokens.
AIFans API serves DeepSeek's latest models through OpenAI- and Anthropic-compatible APIs, pay as you go, no subscription. Priced the same way as DeepSeek (input, cache hits and output billed separately, with peak and off-peak hours), our per-token prices are lower:
DeepSeek-V4.1-Flash (model: deepseek-v4.1-flash)
DeepSeek's own name for it is deepseek-flash. 40% less at peak and off-peak. Model page and code samples
| Item (USD per 1M tokens) | Period | DeepSeek | AIFans API | Difference |
|---|---|---|---|---|
| Input (cache miss) | Peak | $0.3 | $0.1791 | 40% less |
| Input (cache miss) | Off-peak | $0.15 | $0.0896 | 40% less |
| Input (cache hit) | Peak | $0.006 | $0.0036 | 40% less |
| Input (cache hit) | Off-peak | $0.003 | $0.0018 | 40% less |
| Output | Peak | $1.2 | $0.7164 | 40% less |
| Output | Off-peak | $0.6 | $0.3582 | 40% less |
DeepSeek-V4-Pro 0813 (model: deepseek-v4-pro)
DeepSeek's own name for it is deepseek-v4-pro. 49% less at peak, 8% less off-peak. Model page and code samples
| Item (USD per 1M tokens) | Period | DeepSeek | AIFans API | Difference |
|---|---|---|---|---|
| Input (cache miss) | Peak | $1.32 | $0.6716 | 49% less |
| Input (cache miss) | Off-peak | $0.66 | $0.6045 | 8% less |
| Input (cache hit) | Peak | $0.044 | $0.0224 | 49% less |
| Input (cache hit) | Off-peak | $0.022 | $0.0201 | 8% less |
| Output | Peak | $3.96 | $2.0149 | 49% less |
| Output | Off-peak | $1.98 | $1.8134 | 8% less |
Online top-ups carry a payment processing fee of 5.5% (minimum $0.8), added at checkout and shown before you pay; it is not part of your credit.
A worked example: one coding-agent step
Coding agents such as Claude Code resend a long context on every step. Take one request with 200K input tokens (195K of them cache hits) and 1,000 output tokens at peak time on deepseek-v4.1-flash. For 1,000 such steps DeepSeek charges about $3.87 and AIFans API about $2.31. Counting the processing fee on a $100 top-up, ours comes to about $2.44, still 37% less. Your real cost depends on your own usage.
Low prices, no hidden catches
The usual worries about cheap APIs, answered one by one:
The model you ask for
A request for deepseek-v4.1-flash is answered by DeepSeek-V4.1-Flash, and so on for every model: no smaller or different model is swapped in, and nothing is silently downgraded when it is busy. When a model is busy you get a clear 429 that says how many seconds to wait. If a model is replaced or retired, the console tells you in advance.
Honest token counts
Token counts are taken from the model provider's own usage report for each call, with no markup or rounding up, and billed to a millionth of a dollar. The console lists input, cache-hit and output tokens and the cost of every call, and exports them as CSV so you can check.
Cache hits at the cache price
The part of a prompt served from the context cache is billed at the cached-input price, using the provider's own cache-hit count, shown separately for every call. For coding agents most input is cached.
Failed calls are free
Calls that are throttled (429), fail on the provider's side, or are cut off by a provider error are never billed. Only when you disconnect yourself is the part already generated billed (estimated from the text in the rare case that the provider reports no usage).
Your data is not for sale
We do not store your prompts or the model's answers; we keep only what billing needs (time, model, token counts, cost). We do not sell your data and do not use it to train models. To answer a request it passes through our servers in Singapore to the provider that runs the model. See the Privacy Policy.
No hidden fees
No subscription, no minimum spend, no monthly fee. Online top-ups carry a payment processing fee of 5.5% (minimum $0.8), added at checkout and shown before you pay; it is not part of your credit. Price changes are announced in advance and apply only to later usage.
Refundable credit
Paid credit does not expire while your account is open, and unused credit can be refunded (the processing fee is not refundable). Free vouchers expire as stated and are not refundable.
FAQ
Is it the same model as DeepSeek's?
Same model version: DeepSeek-V4.1-Flash, DeepSeek-V4-Pro 0813. Send deepseek-v4.1-flash, deepseek-v4-pro as the model. Thinking is on by default and the thinking parameter switches it off, as on DeepSeek's API.
Do you swap in a smaller model or inflate token counts?
No. Each model name is answered by that model version, never a substitute or a downgrade, and token counts come from the provider's own usage report; every call is listed in the console and can be exported.
Will my data be sold or used for training?
We do not store prompts or answers, we do not sell data and we do not train on it. It passes through our servers in Singapore to the model provider only to answer the request.
Am I charged for failed requests?
No. Throttled requests and provider errors are free; only a stream you cut off yourself is billed for the part already generated.
How do I switch from DeepSeek's API?
Change the base URL to https://api.aifans.fan/v1 (Anthropic API: https://api.aifans.fan), use an API key from AIFans API, and use the model names above. The OpenAI SDK, Claude Code, Cursor, Cline and other tools work unchanged.
When are peak and off-peak hours?
Peak is 01:00–04:00 and 06:00–10:00 UTC on weekdays; all other times are off-peak, and so are weekends, judged by when a request arrives. Unlike us, DeepSeek also treats Chinese public holidays as off-peak, so on such a weekday DeepSeek's off-peak price can be lower than our peak price.
How do I pay?
Prepaid credit, billed on actual token usage, every call itemised in the console. Pay by card or PayPal. Online top-ups carry a payment processing fee of 5.5% (minimum $0.8), added at checkout and shown before you pay; it is not part of your credit. New accounts get $1 of free credit after verifying their email.
Notes
- DeepSeek may change its prices at any time; its website is authoritative. The comparison uses prices checked on 2026-10-16.
- Our prices may change; changes are announced in the console in advance, and the console shows the price that applies.
- In thinking mode the reasoning is billed as output tokens, on both sides.
- DeepSeek is a trademark of its owner. We are not affiliated with, endorsed by or partnered with DeepSeek.