DeepSeek · price per token

DeepSeek V4 Pro pricing

What DeepSeek V4 Pro costs per million tokens: input, output, cache read and cache write, with batch and long-context rates where the provider publishes them. Every figure is the rate Vigil's proxy bills with, read from the provider's own page.

Prices checked 9 Oct 2026 against api-docs.deepseek.com. All shown rates agreed with the page. DeepSeek bills by UTC hour: the peak rate on weekdays 01:00–04:00 and 06:00–10:00 UTC, and half that at every other time. The off-peak figure is on DeepSeek’s page, but the registry holds it as an inference from that rule rather than a read row, so only the peak rate is shown here.

Not shown. A rate (weekdays 00:00–01:00 and 04:00–06:00 and 10:00–24:00 UTC) is in the registry as inferred, on no page we read, so it is not shown. A rate (weekend 00:00–24:00 UTC) is in the registry as inferred, on no page we read, so it is not shown.

DeepSeek V4 Pro on DeepSeek API

Model id deepseek-v4-pro. Verified against the provider's published price; registry row retrieved 2026-09-11 from api-docs.deepseek.com.

AppliesInputOutputCache readCache write
weekdays 01:00–04:00 and 06:00–10:00 UTC$1.32$3.96$0.044—

USD per 1M tokens. A dash means the provider publishes no rate for that dimension, so the proxy declines to price it.

  • Endpoint classes: global.
  • on_demand: the standard rate.
  • No long-context tier: the standard rate applies at any prompt size.

Price a DeepSeek V4 Pro call

DeepSeek V4 Pro on DeepSeek API, standard rate, priced now

Per call$0.0164

Cache writes, batch, endpoint class and monthly volume: AI token calculator

Questions

What does a cache read cost on DeepSeek V4 Pro?
A cache-read token on DeepSeek API costs $0.044 per 1M tokens, against $1.32 for an ordinary input token. An ordinary input token costs 30 times as much. No cache-write rate is published for this model.
Does DeepSeek V4 Pro have a batch price?
No batch rate is published for DeepSeek V4 Pro on DeepSeek API, so the proxy prices every call at the standard rate and does not guess at a discount.

Other models: Claude Opus 5 · Claude Sonnet 5 · Claude Haiku 4.5 · Claude Fable 5.1 · Claude Opus 4.8 · Claude Sonnet 4.6 · GPT-6 Sol · GPT-6 Luna · GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.3 Codex · GPT-4o · GPT-4o mini · Gemini 3.1 Pro · Gemini 3.5 Flash · Gemini 3.8 Flash · DeepSeek Flash · Grok 4.6 · Kimi K3