Token prices(USD/1M tokens)

USD / 1M tokens
Model	Input	Output	Cached input	Source
deepseek-flashDeepSeek	$0.30	$1.20	$0.006	Source Oct 10
deepseek-v4-proDeepSeek	$1.32	$3.96	$0.044	Source Oct 10

Standard API rates. Taxes, tools and subscriptions excluded.

Context

deepseek-flash · Peak-hour rate. Off-peak rates are half these rates; see official UTC hours.

deepseek-v4-pro · Peak-hour rate. Off-peak rates are half these rates; see official UTC hours.

Estimate token cost

Token-only estimate, not an actual task bill. Extra tool calls and retries are excluded.

Value for money →

Frequently asked questions

What unit are AI token prices quoted in?

Prices are US dollars per one million tokens. Input, output and cached input are separate rates. These are API prices, not ChatGPT, Claude or Grok subscription fees.

How is an API request cost calculated?

Multiply each token count by its matching rate, add the amounts, then divide by one million. Cached input replaces part of the input count; it is not counted twice. The calculator uses verified rates and shows unknown when a required rate is unavailable.

Does changing a model change the table order?

On the home comparison table, selecting a model updates only that provider row. Click an input or output column arrow to sort explicitly. A low input price does not necessarily mean a low total cost.

Can I assume every input token is billed at the cached price?

No. A listed cached rate does not guarantee a cache hit. Use cached billing only for input covered by the provider’s cache rules. The calculator requires cached input to be no greater than total input and does not assume a discount when its rate is unknown.

Can the token price change for a long context?

Some published pricing conditions distinguish context lengths or other request settings. Check the context notes and source for the selected model. The calculator uses a recorded long-context rate when available; it does not invent a missing rate.

Is the model with the cheapest input always cheapest overall?

No. Total cost depends on input, output and applicable cached tokens. A model with cheap input but expensive output may cost more for a long answer. Enter your expected token counts in the calculator and compare the rates separately.