DeepSeek's API pricing page confirmed this week that its flat, famously cheap token rates are gone. Effective August 16 at 16:00 UTC, the Hangzhou-based lab replaced single flat prices for DeepSeek-V4-Pro and DeepSeek-V4-Flash with a two-tier peak and off-peak schedule, with increases ranging from roughly 50% to more than 1,100% depending on the model, token type and time of day.

What's changing

Peak hours run 01:00–04:00 UTC and 06:00–10:00 UTC daily, windows that overlap peak business hours in Asia and the European morning. Off-peak rates are set at half the peak price. For V4-Pro, output tokens that previously cost a flat $0.87 per million now cost $3.96 per million at peak and $1.98 per million off-peak — more than a fourfold increase even at the cheaper rate. V4-Flash output climbs from a flat $0.28 per million to $1.32 at peak and $0.66 off-peak. DeepSeek says the shift to dynamic pricing is meant to "allocate resources more reasonably" as demand for its models has strained serving capacity.

A pattern DeepSeek flagged but didn't finalize

The increase isn't a surprise: DeepSeek warned developers on August 6 that a "significant" price change was coming, and when the company took DeepSeek-V4-Pro-0813 out of preview on August 13, its own pricing page already flagged that a hike was imminent without publishing final numbers. This week's update fills in those figures and, notably, introduces time-of-day billing — a first for the company — rather than simply raising a flat rate.

Why it matters

DeepSeek's ultra-low API pricing was a core part of its appeal over the past two years, undercutting Western labs badly enough to trigger price cuts across the industry and driving rapid adoption among developers building cost-sensitive agents and coding tools. Ending flat-rate billing in favor of demand-based pricing brings DeepSeek's economics closer to how cloud compute is typically priced, and it signals the company is prioritizing margin and capacity management over being the cheapest option on the market. For teams that built agent pipelines around DeepSeek's near-zero token costs, the change means re-evaluating architecture and budgets — or shifting latency-tolerant workloads into DeepSeek's new off-peak windows to blunt the impact.