Revised AI benchmarks suggest american models maintain significant lead over chinese competitors

New performance metrics from Artificial Analysis indicate that the perceived competitive threat from Chinese AI models may be overstated. The firm recently updated its testing methodology, incorporating secret tests to prevent models from being specifically optimized for them. Under this new system, the gap between the top open-weight models and the overall leader widened from three points to twelve points. Specifically, the leading open-weight models scored 32 points, while Anthropic's Claude Opus 5.5 achieved 58 points. Sridhar Ramaswamy, CEO of Snowflake, noted the dramatic shift, stating the open-model gap jumped from three to twelve points once benchmarks became private, suggesting previous results might have been skewed by benchmark hacking, a concept also cited by Microsoft CEO Satya Nadella. Furthermore, usage data from OpenRouter, examined by Dax Raad of OpenCode, suggests that the traffic previously attributed to Chinese models could be handled by a small computing cluster costing $6 million to $7 million. This, combined with the fact that over one billion people use ChatGPT monthly, suggests that major providers like OpenAI and Anthropic continue to drive the vast majority of industry infrastructure investment, challenging the narrative of rapid Chinese market takeover.