DeepSeek introduces intensity-based pricing for its API

Summary

Starting August 17, 2026, DeepSeek will implement time-of-day-based API pricing. The hours between 09:00–12:00 and 14:00–18:00 Beijing Time will be defined as 'peak hours.' The off-peak rate will be exactly half the peak-hour price for both the v4-flash and v4-pro models.

Why it matters

Time-based pricing serves as a demand management tool. This move indicates that DeepSeek is approaching capacity limits during business hours in China, its primary market. Given the company's previous strategy of aggressive competition through fixed, low prices, this is a notable strategic shift. It also offers developers a concrete 50% discount for workloads running intermittently or overnight. This could reshape how Chinese startups schedule their inference tasks. The move places DeepSeek in the same pricing strategy camp as Western cloud providers and may set a precedent for other major Chinese LLM providers.

Highlights

• For the v4-flash and v4-pro models, off-peak rates are a fixed 50% discount off peak-hour prices across all token categories (cache hit, cache miss, output).
• The peak-hour window covers only 6 hours of the day (09:00–12:00 and 14:00–18:00 Beijing Time), leaving a wide off-peak band for cost-sensitive workloads.
• Absolute prices remain extremely low: v4-flash output is 9 yuan per million tokens during peak hours and 4.5 yuan per million tokens off-peak. These figures are well below most Western counterparts.

Critical take

The article is merely a bare transcription of a pricing table from IT Home and provides no context. It does not specify whether these prices are new or revised. It also fails to explain whether the change stems from GPU scarcity or a deliberate margin strategy. The treatment of existing contracts remains unclear. Readers are left to speculate about the strategic intent behind what appears on the surface to be a routine repricing.