Frequently asked questions
What does the DeepSeek API cost?
Flash: $0.15 per million input tokens (cache miss) and $0.60 per million output tokens at off-peak rates. Double that during peak hours. V4 Pro is roughly three to four times more expensive. Full table on the pricing page.
When are peak hours?
UTC 01:00–04:00 and 06:00–10:00, Monday to Friday, excluding Chinese public holidays. In Beijing time that is 09:00–12:00 and 14:00–18:00. Everything else is off-peak.
How much do cache hits save?
A cache hit costs 1/50th of a cache miss. Fixed content at the front of your request — system prompt, knowledge base prefix, long document — gets cached across calls.
Flash or V4 Pro?
Flash costs about a third to a quarter of V4 Pro and supports vision input. Use Flash for classification, extraction, translation and everyday chat. Use V4 Pro when the task needs a reasoning chain: complex analysis, code review, professional judgement.
How do I estimate a monthly budget?
Take your daily call volume and average input/output length, then use the cost calculator. Or look at one week of real billing and extrapolate.
Do providers really charge different prices for the same model?
Yes. DeepSeek publishes open weights, so dozens of providers resell it. For V4.1 Flash 1017 we track 45 providers with a price spread of 18x. See the provider comparison.
Is the cheapest provider a good idea?
Not necessarily. The lowest prices usually come with FP4 quantization, which reduces output quality. For classification and extraction that may be fine; for reasoning and code, prefer FP8 or better.
Will prices change?
They have, several times. DeepSeek cut prices in 2024, ran off-peak discounts in 2025, introduced peak-hour doubling in mid-2026 and raised prices in August 2026. We track every change on the price history page.
Is this site official?
No. This is an independent third-party project. We compile public pricing data and build calculators. We do not provide API access, resell DeepSeek services, or represent DeepSeek in any way.