DeepSeek has just rolled out the beta of its V4 AI models, shaking up China’s fierce AI landscape by offering output tokens at about $0.87 per million. This pricing starkly undercuts Anthropic’s leading Fable model, which still sits at $50 per million tokens, setting a new low bar in token costs.
What makes the V4 launch notable is its blend of innovation and familiarity. The upgraded V4-Flash API retains its original design but comes with a retrained boost, ensuring developers won’t have to overhaul their existing setups while benefiting from improved performance. The cost advantage is no coincidence. With U.S. export controls limiting access to certain chips, Chinese AI firms have aggressively optimized for affordability. DeepSeek’s price is less than a fraction of competitors like Z.ai’s GLM-5.2 at $4.40 and Moonshot’s pricier Kimi K3 at $15.
This price war intensifies competition with giants like Alibaba vying for AI dominance in China’s rapidly evolving market. Chinese models now dominate OpenRouter’s top ten developer marketplace slots, with DeepSeek, Moonshot, and Z.ai collectively holding six positions. Despite this, prediction markets still heavily favor Anthropic’s AI leadership through August 2026, suggesting the battle is far from settled.
The shift in pricing dynamics could reshape access and adoption across industries as AI models become more affordable for developers and enterprises alike. At the same time, DeepSeek’s strategic upgrade approach signals confidence that cost leadership and technical continuity will win over users amid a crowded field.
This article is for informational purposes only and does not constitute financial advice.
