DeepSeek V4 Pricing per Million Tokens

Peak hours are 09:00–12:00 and 14:00–18:00 Beijing time (01:00–04:00 and 06:00–10:00 UTC), Monday–Friday, excluding Chinese public holidays. All other hours, weekends, and Chinese public holidays are off-peak.

The current pricing page does not state an effective date for this latest table; the date above records our verification. USD and CNY values are taken from the respective official English and Chinese tables, not converted using an exchange rate.

Official Flash API pricing: Off-peak: $0.15 input / $0.6 output per 1M tokens (cache-miss input); Peak: $0.3 input / $1.2 output per 1M tokens (cache-miss input). Peak hours are 09:00–12:00 and 14:00–18:00 Beijing time (01:00–04:00 and 06:00–10:00 UTC), Monday–Friday, excluding Chinese public holidays. All other hours, weekends, and Chinese public holidays are off-peak. Cached input is billed separately; see the complete pricing table.

Official Pro API pricing: Off-peak: $0.66 input / $1.98 output per 1M tokens (cache-miss input); Peak: $1.32 input / $3.96 output per 1M tokens (cache-miss input). Peak hours are 09:00–12:00 and 14:00–18:00 Beijing time (01:00–04:00 and 06:00–10:00 UTC), Monday–Friday, excluding Chinese public holidays. All other hours, weekends, and Chinese public holidays are off-peak. Cached input is billed separately; see the complete pricing table.

The official DeepSeek page separates input pricing into cache hit and cache miss because repeated context can bill differently from fresh context. If your workload reuses long prompts, system instructions, or prior context effectively, the cost profile changes a lot.

That means the right buyer question is not only 'Flash or Pro'. It is also 'How often does my input qualify as a cache hit'. A team that ignores that split can overestimate or underestimate DeepSeek spend by a wide margin.

Use Flash as the financial default when the workload is broad support chat, repeated structured transforms, agent side tasks, or other traffic where throughput matters more than squeezing out the strongest reasoning on every call.

Use Pro when the prompts are harder, the reasoning path matters, or the downstream cost of a weaker answer is higher than the API savings. The official page does not tell you which one to buy. It gives you the contract so you can choose intentionally.

The clean DeepSeek-first rule is simple: start from the user problem, then map that problem onto the official Pro or Flash pricing curve instead of treating the most expensive model as the default out of habit.

The same official page also shows a shared 1M context length and a maximum output of 384K for both `deepseek-flash` and `deepseek-v4-pro`. That matters because model selection is really a three-part decision: price, capability, and traffic shape.

If one model were cheaper but had a radically smaller context or output ceiling, the tradeoff would look different. The current official table says the main split is economic and operational rather than a smaller context window on Flash.

This site's `/pricing` page is for in-stock DeepSeek Coding Plans only. It is not a mirror of every possible official DeepSeek model or every future contract DeepSeek might document.

That distinction matters. The official docs tell you what DeepSeek itself currently charges and supports. The local `/pricing` page only reflects what inventory is actually available for resale on this site.

Answer six questions before standardizing on a DeepSeek model: how many tokens you send in, how many you expect back, how often cached context repeats, whether the workload is output-heavy, how much concurrency you need, and whether weaker reasoning would create downstream support cost.

If you need the throughput side of that decision, continue with `/guides/deepseek-v4-concurrency-limits`. If you need broader endpoint selection context, compare this page with `/guides/deepseek-openai-vs-anthropic-api-routing` and `/guides/deepseek-v4-pro-vs-flash`.