Global AI Inference Costs Hit Yearly Low Amid Competition from Chinese Models

The cost for businesses to run artificial intelligence models has fallen to a yearly low, driven by a global price war and the rise of low cost Chinese open source tools. According to research from investment bank Jefferies citing Silicon Data, average inference prices ranged between US$1.16 and US$1.18 from August 6 to 8, down from US$2.04 on May 31. This decline reflects an increasing emphasis on cost efficiency across both the US and Chinese tech ecosystems.

US companies are lowering prices to maintain market share and move toward profitability. OpenAI has reduced fees by 80 percent for its lightweight Luna model, while Anthropic has introduced a model with high performance at half the price of its most powerful system. Other US players, including Elon Musk's xAI and Meta, have also released lower cost options.

Chinese AI models, such as those from DeepSeek, Alibaba's Qwen series, and Moonshot AI's Kimi K3, are challenging US leadership. Many of these Chinese models are open source, allowing programmers to modify them. This open source approach is closing the performance gap with closed models faster than expected. In Beijing, the AGI Bar provides free tokens of DeepSeek V4 Flash to customers by self hosting the model on Nvidia workstations.

While some Chinese companies are raising fees as their tools gain global traction, analysts suggest that the trend toward cheaper models is inevitable. This shift may lead to a boom in downstream applications, particularly as enterprises seek AI agents that do not cost more than an employee's annual salary.