A new way of measuring AI models suggests China's threat to OpenAI and Anthropic may have been overblown (www.businessinsider.com)

Recent evaluations by Artificial Analysis reveal that Chinese AI models significantly lag behind American counterparts, particularly when using a revised testing methodology that keeps benchmarks secret. This change has broadened the performance gap between the leading models, increasing the score disparity from just three points to twelve points. Under the new criteria, Anthropic's Claude Opus 5.5 achieved a score of 58, while the best open-weight model from Moonshot managed only 32. This heightened gap underscores the potential robustness of American AI technology and may alleviate investor concerns regarding competition from cheaper Chinese models. Furthermore, analyses highlight that the actual usage of open-weight models remains relatively low, despite their appeal due to lower costs and flexibility. Estimates suggest that a minimal amount of computing power could manage much of the demand for these models, indicating that the substantial investments being funneled into AI infrastructure are likely driven by major players like OpenAI and Anthropic. These insights could reshape current narratives around the AI landscape, suggesting that the perceived threat from open-weight models, particularly those stemming from China, might have been overstated, thereby bolstering the attractiveness of investing in companies like Anthropic as they approach an IPO.