Media365
Analysts said Chinese open-source models were being picked up mostly for their “value for money” compared with other models, according to Luo Liang, a Beijing-based AI analyst at consultancy Intelligent Parameters.
Luo said many users had reported letting the autonomous agent run on its own, only to find it consumed large amounts of tokens – the basic units of text or data processed by AI models that directly affect computing costs – and incurred unexpected service bills.
Chinese open-source AI models – since the emergence of DeepSeek’s high-performance and low-cost V3 and R1 systems – have become known for their competitive pricing compared with US models.
Beijing-based Moonshot AI’s newly released Kimi K2.5, hailed as the strongest open-source AI model so far, costs US$0.58 for every million input tokens and US$3 for output.
