SenseTime targets AI efficiency, cuts token use by 60%

SenseTime (00020.HK) is targeting the high cost of artificial intelligence operations with a new model that reduces token consumption by 60 percent, a move aimed at improving efficiency in a competitive market.

The company announced the official rollout of its new-generation lightweight multimodal agent model, "SenseNova 6.7 Flash-Lite." According to the announcement, the model's architecture achieves a leap in agent capabilities with a smaller parameter size, substantially reducing token consumption during inference, particularly in information search scenarios.

Built on a native multimodal architecture, the model can interpret complex webpage layouts, document structures, and financial charts directly, a key difference from models that require an intermediate visual-to-text conversion layer. This allows it to better handle long-chain complex tasks such as data analysis, in-depth research, and presentation generation. The 60% reduction in token usage is benchmarked against text-only agent models performing similar tasks.

The push for greater efficiency addresses a critical issue in the AI industry: the high cost of deploying large-scale models. By lowering token consumption, SenseTime could make its enterprise solutions more attractive to customers, positioning it to better compete with larger players like OpenAI, Google, and Anthropic. This technological advancement could strengthen SenseTime's competitive advantage and is likely to be viewed positively by investors as a potential driver for revenue growth.

This article is for informational purposes only and does not constitute investment advice.