Chinese AI company DeepSeek has launched a new open‑weight language model called V4‑Flash, introducing it with improved agent capabilities and significantly lower usage costs. The firm cut token prices by 50% and suspended a previously announced dynamic‑pricing mechanism that would have doubled costs during peak hours.
Despite these changes, the V4‑Flash still trails the leading domestic model, Moonshot’s Kimi K3, in overall performance and features.
Why this matters
The price cut and the pause of dynamic pricing indicate that DeepSeek is pursuing market share through lower prices rather than higher revenue per token. This move reflects an intensifying price competition across China’s AI sector, where firms increasingly undercut one another on costs.
At the same time, the Chinese government’s role is contradictory. Beijing — a significant investor in the industry — has publicly warned tech firms against “involution,” meaning counterproductive races to the bottom. Yet the state also subsidizes compute and energy purchases for the sector. Experts argue those subsidies can perpetuate the price war by keeping companies afloat even when their business models are not profitable.
Consequences
In the short term, users benefit from lower token costs and more affordable access to models like V4‑Flash. Over the longer term, sustained price competition risks reducing incentives for investment in quality, safety, and innovation, and could lead to market consolidation favoring firms that receive state support or achieve economies of scale.
In summary: DeepSeek’s launch of V4‑Flash and its 50% token price cut highlight the accelerating AI price war in China, a dynamic shaped both by firm strategies and by government subsidies and warnings about destructive competition.
