DeepSeek Makes Devastating Price Cut Permanent: Now 17 Times Cheaper Than OpenAI Per Million Tokens DeepSeek is done running promotions. The Chinese AI lab has made its 75% price cut on V4-Pro permanent, settling the model at 44 cents per million input tokens and 87 cents per million output tokens. OpenAI’s GPT-5.5 costs $2.50 input and $15 output per million tokens. That is a 5x gap on input and a 17x gap on output. This is not a temporary competitive move. It is a pricing floor being set. The lighter DeepSeek V4-Flash goes lower still, at 14 cents input and 28 cents output. DeepSeek’s technical paper explains how these prices work without reporting losses: a hybrid sparse-attention architecture that activates approximately 27% of per-token compute at long contexts compared to its predecessor, with 10% of the memory requirements. The company engineered its way to these numbers. They reflect real efficiency, not subsidized market capture. What making the cut permanent signals is the important part. DeepSeek is not testing whether developers will switch. It is establishing the cost basis for the Chinese open-weight model tier as a durable market reality. Every enterprise team managing high-volume AI pipelines now faces a genuine financial question about whether the governance, safety guarantees, and reliability of Western frontier models is worth the price differential. For Anthropic and OpenAI, which have both raised prices significantly in 2026, the gap has never been wider. The premium they charge is real. Whether it is worth 17 times more is a calculation that procurement teams are actively running right now. Keywords: DeepSeek price cut permanent, DeepSeek vs OpenAI cost, Chinese AI cheaper, AI inference pricing 2026