When does DeepSeek's peak pricing start?

DeepSeek's peak/off-peak API pricing takes effect when V4 officially launches in mid-July 2026. Users on the API get an email 24 hours before their meter switches to the new rates, so teams have time to shift jobs to off-peak windows.

What are DeepSeek's peak hours?

DeepSeek's peak windows are 9:00 to 12:00 and 14:00 to 18:00 Beijing time, seven days a week. Outside those two windows the API bills at the reduced off-peak rates DeepSeek set in May. There is no separate weekend schedule.

Which DeepSeek models are affected?

DeepSeek's peak/off-peak split applies to V4 Pro and V4 Flash — both the flagship reasoning tier and the lightweight tier. During peak both output and cache-miss input token rates double; off-peak rates stay at the May cuts.

How can teams avoid the DeepSeek peak surcharge?

Move batch workloads — offline evaluations, document ingest, nightly code review — outside DeepSeek's peak windows. Run before 9am Beijing, between 12:00 and 14:00, or after 18:00. Real-time user traffic still pays peak but rarely dominates cost.

DeepSeek · 2026-06-29 · major

DeepSeek V4 gets peak-hour pricing — API doubles 9am–12pm and 2pm–6pm Beijing time

DeepSeek notified users on June 29 that when V4 ships mid-July, API rates for V4 Pro and V4 Flash will double during two Beijing peak windows (9–12 and 14–18) to relieve compute congestion. Off-peak rates hold at the May reductions.

First large LLM API to introduce time-of-day pricing.

Quick facts

Maker	DeepSeek
Effective	Mid-July 2026, at V4 official launch
Peak windows	9:00–12:00 and 14:00–18:00 Beijing time
Peak multiplier	2× the off-peak rate
Affected models	V4 Pro and V4 Flash
Notification	Email 24 hours before rate changes

Pricing

V4 Pro output — off-peak	¥6 / 1M tokens
V4 Pro output — peak · 2× multiplier	¥12 / 1M tokens
V4 Flash output — off-peak	¥2 / 1M tokens
V4 Flash output — peak · 2× multiplier	¥4 / 1M tokens

source ↗

What is it?

DeepSeek V4 will bill differently depending on when a request lands. Two Beijing time windows — 9 AM to 12 PM and 2 PM to 6 PM — charge double the off-peak rate on both V4 Pro and V4 Flash. Off-peak rates hold at the reduced levels DeepSeek set in May.

How does it work?

The API splits the day into peak and off-peak blocks in Beijing local time. During the two peak windows the meter charges 2× for output and cache-miss input tokens; cache-hit input is smaller but also doubles. Users get an email 24 hours before their rates change so teams can reschedule batch jobs.

Why does it matter?

Time-of-day pricing is a first for a major LLM API and mirrors how power grids handle demand. For DeepSeek customers running heavy nightly workloads — code review, document ingest, evaluations — the savings from moving those outside the two Beijing windows are real. Expect other providers to notice.

Who is it for?

Teams running production DeepSeek workloads

Frequently asked questions

When does DeepSeek's peak pricing start?: DeepSeek's peak/off-peak API pricing takes effect when V4 officially launches in mid-July 2026. Users on the API get an email 24 hours before their meter switches to the new rates, so teams have time to shift jobs to off-peak windows.
What are DeepSeek's peak hours?: DeepSeek's peak windows are 9:00 to 12:00 and 14:00 to 18:00 Beijing time, seven days a week. Outside those two windows the API bills at the reduced off-peak rates DeepSeek set in May. There is no separate weekend schedule.
Which DeepSeek models are affected?: DeepSeek's peak/off-peak split applies to V4 Pro and V4 Flash — both the flagship reasoning tier and the lightweight tier. During peak both output and cache-miss input token rates double; off-peak rates stay at the May cuts.
How can teams avoid the DeepSeek peak surcharge?: Move batch workloads — offline evaluations, document ingest, nightly code review — outside DeepSeek's peak windows. Run before 9am Beijing, between 12:00 and 14:00, or after 18:00. Real-time user traffic still pays peak but rarely dominates cost.

Try it

Schedule V4 batch jobs outside Beijing 9–12 and 14–18 windows