OpenAI Just Slashed Prices! DeepSeek Immediately Fires Back with '4x Cheaper Cabbage Pricing' — Foreign Media: The Price War Has Been Shattered

China's AI sector has reignited its price war. Just days after OpenAI announced massive price reductions—up to 80%—for its 'GPT-5.6 Luna' model, Chinese company DeepSeek responded within hours by unveiling its new flagship model, 'V4-Flash,' priced even lower. At just $0.28 per million output tokens, foreign media outlet Wccftech remarked, 'China's AI price war counterattack arrived faster than expected.'

According to the report, OpenAI significantly adjusted the pricing of its GPT-5.6 Luna model. The revised rates are as follows: input costs dropped from $1.00 to $0.20 per million tokens, while output costs fell from $6.00 to $1.20 per million tokens. OpenAI stated this drastic reduction was possible due to further optimization of the model architecture, resulting in improved computational efficiency. However, most analysts believe this move was the first response to the aggressive pricing strategies already deployed by Chinese AI firms.

Even Lower Than OpenAI? DeepSeek's Output Cost Is Just $0.28

The report highlights that mere hours after OpenAI's price drop announcement, DeepSeek launched its latest model, V4-Flash. Notably, this model has only 284 billion parameters, yet the company claims its overall performance approaches that of the Opus 4.8 model developed by U.S. firm Anthropic, which boasts over a trillion parameters—drawing significant market attention.

What shocked the market even more was DeepSeek's direct pricing assault on OpenAI. The V4-Flash is priced at $0.14 per million input tokens and $0.28 per million output tokens. Despite OpenAI's steep discounts, DeepSeek has nearly erased its competitor's pricing advantage.

Directly Distilling U.S. AI? Chinese Firm 'Moonshot' Buys 20,000 NVIDIA Chips

On another front, Bloomberg reported that the rapidly rising Chinese AI startup 'Moonshot AI' has acquired 20,000 NVIDIA H200 chips via Alibaba, dramatically boosting its model training capacity.

However, Moonshot has also faced technical controversies. Both Anthropic and U.S. government officials have accused its Kimi K3 model of using 'distillation' to extract AI capabilities from Anthropic's Fable model. Additionally, according to a recent allegation by a U.S. official, Moonshot not only secretly possesses NVIDIA's GB300 AI servers but also obtained additional GB300 computing power through Thailand, violating export controls.

Does the U.S. Have Competition in Open-Source AI? A Startup Rises to the Challenge

The report notes that in the battle for open-source AI models between the U.S. and China, American AI startup Thinking Machines has recently released an open-weight model called 'Inkling-Small.'

According to benchmark analysis by Artificial Analysis, DeepSeek's V4 Flash leads with a score of 50%. Inkling-Small scored 40%, slightly behind the new DeepSeek version but comparable to the older V4 Flash model.