DeepSeek V4 Pro tops global bang-for-buck ranking after 75% price cut

Price cut and ranking

Hangzhou-based start-up DeepSeek (Chinese name not given in the source) has moved to the front of the cost-efficiency race in generative AI after making a 75% promotional price cut on its flagship V4 Pro permanent. It has been reported that third‑party benchmarker Artificial Analysis ranks the V4 Pro among the world’s best on an “intelligence‑per‑dollar” basis — a measure that has gained traction as compute costs surge globally. How did it pull this off? By undercutting rivals on API costs while claiming competitive performance.

Numbers that matter

DeepSeek’s published API prices reportedly now stand at US$0.0036 per 1 million cached input tokens and US$0.87 per 1 million output tokens. According to Artificial Analysis’ composite Intelligence Index, running the benchmark on V4 Pro costs about US$268 — roughly 12 times cheaper than OpenAI’s GPT‑5.5 and 19 times cheaper than Anthropic’s Claude Opus 4.7 for the same workload, the report says. The V4 generation also includes a lighter V4 Flash variant, positioning the company to target both high‑throughput and lower‑latency use cases.

Context and implications

This move highlights a distinct Chinese playbook in the global AI race: aggressive price competition rather than premium pricing for marginal capability gains. It has been reported that the pricing strategy is partly a response to a global compute supply crunch and higher costs for top‑end models. Geopolitics also looms in the background — US export controls on advanced chips and other trade frictions have complicated access to cutting‑edge compute for some Chinese firms, making cost and software efficiency more salient competitive levers.

What comes next?

Lower prices could accelerate adoption among startups and enterprises worldwide that are sensitive to run costs, but durability remains a question. Will DeepSeek sustain quality and trust at scale? And how will Western incumbents respond — by cutting prices, optimizing models, or leaning on proprietary capabilities and ecosystem lock‑ins? For now, DeepSeek’s permanent cut has reset expectations about the economics of deploying large language models.