DeepSeek Adjusts API Pricing Upward by Up to 1,100% and Implements Peak Usage Rates

Hangzhou-headquartered artificial intelligence startup DeepSeek is enacting substantial cost increases for software developers using its Application Programming Interfaces (APIs). Beginning August 16 at 16:00 UTC, the price to run its V4-Flash and V4-Pro artificial intelligence models will jump anywhere from 50 percent to over 1,100 percent, based on token categories, specific models, and the time of day requests are processed, as first reported by Yahoo Finance.

Alongside the base price hikes, DeepSeek is introducing a time-segmented billing framework to manage server traffic. Under the newly established structure, peak usage times are designated from 01:00 to 04:00 UTC and from 06:00 to 10:00 UTC. During non-peak windows, customers will be charged exactly 50 percent of the peak rates, according to the startup's official pricing schedule.

For output generation on the V4-Flash model, developers will now pay $1.32 per million tokens during peak hours and $0.66 per million tokens off-peak. This represents a marked jump from the previous uniform charge of $0.28 per million tokens. Meanwhile, the higher-tier V4-Pro model will see output prices rise from $0.87 per million tokens to $3.96 per million during peak windows and $1.98 per million in off-peak periods.

Data input costs are escalating as well. For cache-miss input processing on V4-Flash, peak prices will rise from $0.14 per million tokens to $0.44 per million. For V4-Pro, cache-miss input rates during peak hours will shift to $1.32 per million tokens, up from $0.435 per million tokens previously.

Statements from DeepSeek's pricing materials, cited by Fortune, indicate that these alterations are designed "to allocate resources more reasonably." The dual-rate structure seeks to incentivize engineers to redirect automated workloads away from high-density windows to alleviate compute bottlenecks. Despite these sharp rate hikes, DeepSeek's pricing structure remains comparatively frugal compared to mainstream Western rivals. For instance, Anthropic charges $50 per million output tokens for its Fable 5 model, according to Fortune.

The dramatic recalibration underscores a swift evolution in DeepSeek's go-to-market and monetization strategy. When the company launched V4-Flash earlier this month, research entity Artificial Analysis rated it as the most economical prominent AI model globally, estimating its operational cost at roughly three cents per benchmark evaluation. Prior to this, DeepSeek had maintained a 75 percent introductory promotional markdown on V4-Pro through May 5.

While the company signaled last week that rate increases were imminent without disclosing precise figures, the tactical shift arrives during a period of massive corporate expansion. DeepSeek has recently begun laying preliminary groundwork for an initial public offering (IPO), following the completion of its debut institutional investment round that secured more than $7 billion in fresh capital.