The stellar Chinese artificial intelligence model DeepSeek further intensifies the price war despite the pressure on computing power costs in the AI industry, offering a 75 percent price slash.
After launching the newest model V4 last Friday, DeepSeek unveiled a promotion of 25 percent of its original price for the high-level DeepSeek-V4-Pro model, bringing its input price from 1 yuan (HK$1.15) to 0.25 yuan per million tokens - the lowest among global prominent AI models.
DeepSeek's aggressive price cuts - building on its already low costs - are rattling the AI industry, but successfully winning over users, as seen by a nearly four-time jump in its V4-Pro usage.
For comparison, the weighted average input price of the newly launched GPT-5.5 Pro is US$30 (HK$235) per million tokens, representing more than 700 times the price difference between the two models.
Other prevailing AI models like Anthropic Claude Opus and Gemini 3.1 Pro offer output prices ranging from US$12 to US$25 per million tokens, far higher than 6 yuan of DeepSeek-V4-Pro and 2 yuan of its low-cost model DeepSeek-V4-Flash.
For domestic peers, Alibaba (9988)'s Qwen-3.6-Plus features an input price of 2 yuan per million tokens, while MiniMax (0100) offers 2.1 yuan per million tokens.
Zhipu (2513) previously even hiked prices by 10 percent after it released its latest open-source model on rising demand, charging 6 yuan per million tokens.
Some analysts said that the low-cost AI model would further help China to export its tokens, based on its massive reserves of cheap electricity and computing power.
Chinese large language model companies accounted for nearly 36 percent of the total market in terms of token consumption for last week, led by DeepSeek, Moonshot, MiniMax, Zhipu, and Tencent (0700), according to popular AI model marketplace OpenRouter.
𝗙𝗼𝗹𝗹𝗼𝘄 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝘂𝗻𝗶𝘁𝘆 𝗼𝗳 𝗧𝗵𝗲 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱↓
