DeepSeek V4 Flash Undercuts OpenAI as China’s AI Price War Intensifies

DeepSeek has responded to OpenAI’s latest price cuts with a cheaper version of its V4 Flash model, increasing pressure on US AI companies as competition over model pricing, performance, and computing power continues to grow.

OpenAI recently reduced the price of GPT-5.6 Luna by as much as 80 percent, cutting input token pricing to $0.20 per million tokens and output pricing to $1.20 per million tokens. The company linked the lower prices to architectural improvements that reduced operating costs.

DeepSeek followed within hours by launching V4 Flash 0731, a refreshed model with 284 billion parameters that reportedly delivers performance close to Anthropic’s Opus 4.8. DeepSeek priced input tokens at $0.14 per million and output tokens at $0.28 per million, giving it a major price advantage over OpenAI.

Moonshot Expands Its AI Training Capacity

Bloomberg reports that Moonshot has secured access to a cluster of 20,000 Nvidia H200 GPUs through Alibaba, which gives the Chinese AI company far more computing power for training future models.

Moonshot has also faced claims that it used knowledge distillation while developing Kimi K3, although the company has not publicly confirmed those allegations. US officials have also raised concerns about Chinese AI labs accessing advanced Nvidia hardware through other countries.

Meanwhile, Thinking Machines has introduced Inkling-Small, an open-weight model with 276 billion parameters. DeepSeek V4 Flash 0731 currently leads the Artificial Analysis Intelligence Index with a score of 50 percent, while Inkling-Small follows at 40 percent.

The latest moves show that Chinese AI companies are competing aggressively on both price and performance, while US labs face growing pressure to lower costs without falling behind on model quality.