DeepSeek V4-Pro makes its 75% API price cut permanent

DeepSeek has moved a temporary promotion into its baseline API pricing, putting another hard number under the cost pressure facing large model providers. In a May 22, 2026 post on X, the company said V4-Pro’s discounted price will remain in place after the offer period, with the attached image stating: “DeepSeek-V4-Pro 75% OFF - Now Permanent.”

“We are making our discount permanent!”

The concrete change is a fourfold price cut. The image attached to the post lists cache-hit input at $0.003625, down from $0.0145; cache-miss input at $0.435, down from $1.74; and output at $0.87, down from $3.48. The quoted earlier post had extended the V4-Pro discount until May 31, 2026 at 15:59 UTC. The new post turns that deadline into a permanent pricing reset rather than another short promotion.

That matters because the next phase of LLM competition is not only about benchmark rankings. API economics decide whether developers can run coding agents, research assistants, and long-context workflows repeatedly enough to be useful. DeepSeek’s official account is typically used for model releases, API changes, open-weight availability, and pricing updates, so this post functions as a direct pricing signal to developers already testing V4-Pro against proprietary and open model alternatives.

The practical test now shifts to capacity. A low token price helps only if throughput, latency, and rate limits hold up under production load. If V4-Pro can sustain the new pricing without frequent overload errors, it may force buyers to reprice what they expect from higher-cost model APIs. Watch for responses from model routers, enterprise AI platforms, and rival labs that sell coding or agent workloads by token usage. The source post is available on DeepSeek’s X account.