DeepSeek ships V4-Pro and raises API prices by up to 1,100%

DeepSeek moved V4-Pro out of preview as DeepSeek-V4-Pro-0813, available across its app, web interface and API, with the general-availability version centred on agentic capability — using tools, executing code and completing multi-step workflows without human intervention. Alongside the launch it raised API prices across the V4 family by between 50% and 1,100%, varying by model, token type and time of use, effective 16:00 UTC on 16 August. V4-Pro output tokens go from a flat $0.87 per million to $3.96 at peak hours; DeepSeek is also introducing peak and off-peak billing, with off-peak set at half the peak rate.

Why it matters

The company that defined the price floor just raised it — and did so a fortnight after shipping V4-Flash as the cheapest major model to run. Read the two moves together and the strategy is segmentation rather than retreat: cheap inference stays available at the Flash tier while the agentic tier is priced like a product, and time-of-day billing treats DeepSeek's GPUs as a scarce resource to be load-balanced rather than a loss leader. That undercuts the assumption that has shaped Western pricing all year — that Chinese labs would keep driving costs toward zero — at the exact moment OpenAI and Anthropic are cutting prices under pressure from those same rivals. The uncomfortable reading for buyers is that nobody's per-token economics work at the old prices, including the lab everyone cited as proof they could.

What to watch

Whether the increases stick after 16 August or get walked back, whether Western labs stop cutting once the floor lifts, and how much of DeepSeek's agent traffic moves to off-peak windows.