Back
DeepSeek flags relatively large across-the-board API price increase

DeepSeek flags relatively large across-the-board API price increase

The August 6 notice came as China's weekly AI token usage reached 28.13 trillion and after mid-July reports of peak-hour pricing changes, while DeepSeek still had not disclosed the size or timing of the broader increase.

Fact Check
The core claim—that on August 6, 2026 DeepSeek flagged a relatively large across-the-board API price increase without disclosing the size or exact timing—is directly corroborated by PANews (citing Cailianshe) and Eastmoney's Cailianshe-sourced roundup, both quoting the announcement. Bloomingbit corroborates the sharp price-rise framing. BlockBeats and the Securities Times report confirm the mid-July peak/off-peak pricing change context and that current unified pricing remains, retaining a wide cost gap versus global rivals as the claim states. All claim components are supported.
Summary

DeepSeek said on August 6 that it will soon raise API pricing across the board and warned that "a relatively large increase is expected," without disclosing the percentage change or effective date. The move comes as China's weekly large-model token consumption reached 28.13 trillion from July 27 to August 2, extending the country's global lead to 14 straight weeks and adding pressure on providers to refine pricing as compute shortages emerge during peak periods. DeepSeek's announcement followed mid-July reports of peak-valley pricing, while CCTV Finance also reported that rates during 9 a.m. to 12 p.m. and 2 p.m. to 6 p.m. could be set at twice normal levels; earlier coverage noted DeepSeek's official pricing page was still showing a uniform schedule at the time. Even with higher prices, DeepSeek appeared to retain a wide cost gap versus rivals, while other providers including Moonshot AI's Kimi were also moving beyond simple usage-based billing through subscription plans priced from 39 yuan to 559 yuan.

Terms & Concepts
  • API: Application programming interface, the service layer developers use to call an AI model and pay based on usage or another pricing plan.
  • Token: A unit of text processing used to measure AI model input and output consumption for billing and capacity tracking.
  • Peak-valley pricing: A pricing model that charges higher rates during high-demand periods and lower rates during off-peak periods to manage limited computing resources.