AI index v5 widens US-China gap to 12 points, sparking benchmark debate

Artificial Analysis published Intelligence Index v5, and the measured gap between the top US model and the leading Chinese model widened from 3 points to 12 points, according to aitimes.com. In the new index, Anthropic's Claude Opus 5.5 scored 58, while Moonshot AI's Kimi K3.1 Pro scored 46. The highest-scoring open-weight model reached only 32 points.

Google's Gemini 4 Carbon later took first place with 59 points. Artificial Analysis also raised the share of the index built on private questions and answers from 45% to 60%, a change that reshaped the rankings.

Questions over benchmark trust

The revision drew challenges from industry figures. Snowflake CEO Sridhar Ramaswamy questioned on X whether open models were really catching up or whether test questions had leaked.

Trillium Labs co-founder Nathan Lambert said on his blog that the earlier index seems to have rated open-weight models too favorably, but called talk of the collapse of open models greatly exaggerated.

Separately, OpenCode co-founder Dax posted on X on October 27, 2026 an estimate that most of DeepSeek's daily traffic through OpenRouter could be served by a single NVIDIA GB300 rack. Before the v5 revision, expectations had spread that cheap Chinese open-weight models would threaten OpenAI's and Anthropic's business models.