DeepSeek slashes AI model costs 50% amid China price war

Chinese AI giant DeepSeek has released V4-Flash, an open-weight model featuring enhanced agent capabilities, as part of an aggressive cost-cutting strategy that underscores China's intensifying AI price competition. The model launch is accompanied by a 50% reduction in token costs — the pricing unit for AI inference — and a pause on a previously announced dynamic-pricing mechanism that would have doubled costs during peak usage hours. This combination suggests DeepSeek is deliberately sacrificing near-term margins to capture market share. While V4-Flash represents an advance in agent capabilities (features that allow AI models to take autonomous actions), it still trails Moonshot's Kimi K3, which remains the leading Chinese AI model by capability. The price war reflects a broader dynamic China calls "involution" — a self-destructive race to the bottom that has already ravaged profitability in other sectors such as solar panels and electric vehicles. Beijing has publicly warned tech firms against involution, but government policy presents a paradox: the state subsidizes both compute purchases and energy costs for the industry, a practice experts say actively enables the unprofitable competition by keeping marginal companies alive. DeepSeek's latest pricing move suggests that competitive market pressures are dominating strategic considerations, even as official guidance discourages such races.