recently, china’s leading ai company deepseek officially unveiled a new api pricing strategy for its deepseek-v4-pro large model—extending the originally time-limited 25% discount, which was set to expire on may 31, into a permanent pricing structure, effectively slashing overall costs by 75%. following the adjustment, the standard pricing is as follows: cached input at 0.025 yuan per million tokens, non-cached input at 3 yuan per million tokens, and output at 6 yuan per million tokens.
according to the latest benchmark report released by artificial analysis, a globally authoritative ai evaluation organization, deepseek-v4-pro has claimed the top spot in global large-model cost-effectiveness, with a comprehensive cost of just 268 usd per ai intelligence index test run. by comparison, the equivalent tasks performed by openai’s gpt-5.5 and anthropic’s claude opus incur costs 12 times and 19 times higher, respectively. among today’s global top-ranked high‑cost‑performance models, more than half are from china, with deepseek-v4-pro standing out as a benchmark for both technological innovation and commercial success.
in terms of performance, v4-pro has made no compromises despite the price reduction—its agentic coding capabilities have broken open-source model records, becoming deepseek’s primary framework for agent-based programming, delivering results that surpass sonnet 4.5. moreover, in critical areas such as mathematical reasoning, stem proficiency, and algorithmic competition‑style code generation, it comprehensively outperforms all publicly evaluated open-source competitors, reaching a level comparable to the very best closed-source models. this dual-track strategy—maintaining uncompromised performance while achieving dramatic cost reductions—is accelerating the transition of ai technology from an elite tool to widespread, scalable applications.
