DeepSeek hikes V4 API prices as much as 1,100%, ending the cheap-agent era

The Hangzhou lab is switching to peak/off-peak billing on August 16, with V4-Flash output rising from $0.28 to $1.32 per million tokens and cache-hit reads jumping roughly sixfold — the change that hits agent pipelines hardest.

DeepSeek is raising API prices on its V4 model family by roughly 50% to more than 1,100%, depending on model, token type, and time of day, according to Reuters. The Hangzhou lab published the change on August 13; new rates take effect August 16 at 16:00 UTC, per its own documentation. The company framed the shift as “revising and adjusting the pricing to allocate resources more reasonably.”

The headline numbers, via Bloomberg: V4-Flash output moves from a flat $0.28 per million tokens to $1.32 at peak and $0.66 off-peak. V4-Pro output goes from $0.87 to $3.96 peak and $1.98 off-peak. Peak windows are 01:00–04:00 and 06:00–10:00 UTC, which The Decoder notes maps cleanly onto the Chinese workday.

The sharpest move in relative terms isn’t in the marquee output rates. It’s in cache-hit reads. V4-Pro’s cached input pricing rises from $0.003625 to $0.022 off-peak and $0.044 at peak per million tokens, a roughly sixfold jump. Cached reads had been about one-hundred-twentieth of the regular input rate; they’re now closer to one-thirtieth. The Decoder flags this as the most expensive part of the switch for agent developers, whose pipelines lean hardest on repeatedly re-reading long context windows.

DeepSeek paired the price sheet with a refreshed build, V4-Pro-0813, keeping the same parameter count and one-million-token context and adding OpenAI Responses API support with Codex integration. It also open-sourced the agent runtime as DeepSeek Harness.

Context matters here. Even at the new peak rates, DeepSeek sits nowhere near Anthropic’s Fable 5, which Bloomberg pegs at $50 per million output tokens. And this hike only partially unwinds the May price cut that helped cement DeepSeek’s reputation as the cheap-agent substrate.

Bloomberg reports the lab is preparing for a potential IPO as soon as this year. The pricing page’s language about resource allocation reads differently in that light. The commodity-scale loss-leader phase of Chinese frontier inference is being repriced for a book that has to look profitable to an underwriter.

DeepSeek hikes V4 API prices as much as 1,100%, ending the cheap-agent era

Sources