Google Gemini 4 Carbon leads latest blind-test rankings
Google’s Gemini 4 Carbon has debuted at the top of the Arena blind-test leaderboard. Meanwhile, Chinese AI models are showing consistent performance across development and video generation categories.
The latest Arena blind-test results for mid-November 2026 indicate a shifting landscape in large language model performance. Google’s newly released Gemini 4 Carbon (High) secured the top spot with an ELO score of 1526, narrowly edging out its sibling model, Gemini 4 Argon. The results reflect user subjective preferences gathered through anonymous head-to-head testing.
Anthropic continues to maintain a significant presence in the developer-focused segments, with its Claude series dominating the front-end coding leaderboards. The data highlights a highly competitive environment where established players face pressure from rapidly updated model iterations.
Chinese models are increasingly penetrating global leaderboards, particularly in the video generation and software development sectors. Notable performers include Moonshot AI’s Kimi K3.1 Pro, which entered the top 15 in the front-end category, and offerings from ByteDance and Alibaba, which have secured stable positions in the AI video generation space. These developments suggest that regional AI developers are narrowing the gap in specialized multimodal tasks.