Guide

DeepSeek V4 Peak Hours Are Live: Move Batch Jobs or Pay Double

By Istar editors
At a glance

DeepSeek V4 API rates double during two weekday UTC windows. Use the published clock and current token table to schedule batch work.

DeepSeek now bills V4 API tokens at peak and off-peak rates. Peak hours are 01:00–04:00 UTC and 06:00–10:00 UTC, Monday through Friday. Every other hour, including the whole UTC weekend, is off-peak. Each listed off-peak rate is half its matching peak rate.

For deepseek-v4-pro, output costs $1.98 per million tokens off-peak and $3.96 peak. Cache-miss input is $0.66 off-peak and $1.32 peak. Moving the same token load with the same cache composition to an off-peak window halves that token bill.

DeepSeek announced the change on 13 August 2026 and put it into effect at 16:00 UTC on 16 August 2026. The prices and schedule below were checked on the live Models & Pricing page on 28 August 2026.

The schedule DeepSeek published

UTC, Monday–Friday Beijing time, same weekday Rate
01:00–04:00 09:00–12:00 Peak
04:00–06:00 12:00–14:00 Off-peak
06:00–10:00 14:00–18:00 Peak
10:00–01:00 next day 18:00–09:00 next day Off-peak
Saturday and Sunday Saturday and Sunday Off-peak

Use an explicit timezone in the scheduler. A host-local cron can move when daylight-saving time changes even though DeepSeek’s clock remains UTC. For batch work, leave a small buffer around 01:00, 04:00, 06:00, and 10:00 UTC; the pricing page does not explain how a request crossing a boundary is classified.

The weekend statement is based on the current English page’s “Monday through Friday” qualifier. Older cached versions of the pricing page described the planned peak windows differently. Use the live table, not an old screenshot or search snippet.

Current V4 Flash and V4 Pro prices

U.S. dollars per 1 million tokens:

Model Billing item Off-peak Peak
deepseek-v4-flash Input, cache hit $0.007 $0.014
deepseek-v4-flash Input, cache miss $0.22 $0.44
deepseek-v4-flash Output $0.66 $1.32
deepseek-v4-pro Input, cache hit $0.022 $0.044
deepseek-v4-pro Input, cache miss $0.66 $1.32
deepseek-v4-pro Output $1.98 $3.96

The Chinese pricing page publishes a separate RMB table. Use that table for RMB billing rather than converting these dollar figures with a live exchange rate.

DeepSeek also lists deepseek-v4-flash-vision-exp at the same dollar token rates as V4 Flash, with image inputs converted to tokens based on dimensions. It is an experimental vision model released after the V4-Pro pricing announcement, so this scheduling guide does not use it in the cost examples.

What moving one batch saves

Suppose one deepseek-v4-pro batch produces:

  • 10 million cache-miss input tokens
  • 10 million output tokens
  • no cache-hit input in this simplified example

At peak, the token cost is:

10 × $1.32 + 10 × $3.96 = $52.80

Off-peak, the same measured usage is:

10 × $0.66 + 10 × $1.98 = $26.40

The $26.40 difference comes from timing only. If the prompt changes, output length changes, or cache-hit share changes, compare the actual usage categories instead of calling the jobs identical.

The same rule applies to V4 Flash: each off-peak line is one-half of its paired peak line. The lower absolute price may make scheduling less important for a small job, but the percentage difference is the same.

Which work should move

Good candidates are jobs whose start time is flexible and whose result does not block a person:

  • nightly classification or extraction
  • bulk rewriting and localization
  • offline evaluations
  • repository indexing and summarization
  • report generation for the next workday

Interactive agents and customer-facing requests may not be worth delaying. Route those for latency and business need, then use the cheaper window for the backlog.

For a Beijing-time team, the easiest weekday off-peak gap is 12:00–14:00, with a longer window after 18:00 and before 09:00. Weekends are entirely off-peak under the current rule. For any other region, derive the schedule from UTC and test it across the next daylight-saving transition.

Balance order does not change with the clock

DeepSeek calculates the expense as token usage multiplied by the applicable model rate. When an account has both granted and topped-up balance, the pricing page says granted balance is deducted first, followed by topped-up balance.

That ordering does not make granted credits a different peak/off-peak price class. Both balances pay the rate attached to the measured request. Track token categories and time window separately from the source of funds.

DeepSeek reserves the right to adjust product prices and recommends checking the pricing page before topping up. Store the live table alongside any material cost estimate; this is a scheduling policy, not a permanent price guarantee.

This article expands the batch-routing example in when not to buy top models without changing that guide’s broader model-selection decision.

Sources