Chinese large models make strong showing in weekly AI rankings

The Arena platform released its weekly AI model rankings for the third week of 2026, highlighting significant progress by several Chinese large language models. The newly released GLM-5.3-Max from Zhipu AI made an immediate impact, securing a position in the top 15 of the general ranking at 13th place with an ELO score of 1487. Meanwhile, Kimi-K3-Max from Moonshot AI rose from 12th to 10th, marking its first entry into the top 10. Other notable entries included Muse-Spark-1.1, which climbed to 8th, and new models such as GPT-5.5-Instant and Claude-Opus-4-8, which debuted at 28th and 29th respectively. Qwen3.8-Max from Alibaba saw a decline in ranking, dropping from 8th to 19th. In the code ranking, Claude-Opus-4-7-High took the lead with an ELO score of 1552, while GLM-5.3-Max entered the top 10 at 10th place with an ELO of 1531. The new model Grok-4.20-Beta1 debuted at 20th, and other new entries included Claude-Sonnet-4-5-20250929-High-32K and GPT-5.5-High. Qwen3.8-Max fell from 13th to 25th in this category. In the front-end development ranking, Claude-Opus-5-Max maintained its lead with an ELO of 1691, while Kimi-K3-Max and Qwen3.8-Max occupied second and third places with scores of 1674 and 1669 respectively. GLM-5.3-Max debuted at 8th, and Qwen3.8-27B entered the top 10 at 9th. These results indicate growing competitiveness among Chinese models in this domain. The AI image generation ranking remained stable, with GPT-Image-2 (Medium) leading at 1381 points. Seedream-5.0-Pro and Qwen-Image-3.0-Pro maintained positions at 8th and 9th respectively. Hunyuan-Image-3.0 rose to 29th. In the AI video ranking, ByteDance’s Dreamina-Seedance-2.5-720P debuted at 4th with an ELO of 1477, following its predecessor, Dreamina-Seedance-2.0-720P, which ranked third. Gemini-Omni-Flash remained first with 1512 points, and Flux-3-Video came in second. In the agent ranking, Anthropic models dominated, with Claude Opus 5 (High) leading at 12.47% net improvement. Kimi-K3-Max from Moonshot rose to fourth with a net improvement of 10.41% and a task confirmation success rate of 17.97%. New entries included DeepSeek V4 Pro (High) (0813) at 14th and Qwen3.8-Max at 15th. GLM-5.2 (Max) from Zhipu AI ranked 17th with a net improvement of 5.82%, while DeepSeek V4 Flash (High) (20260731) placed 20th with a net improvement of 3.99%. Overall, the week’s rankings showcased strong performance from Chinese models across multiple categories, with several new entries making notable impacts. As more Chinese large models are expected to release new versions, the competition in the rankings is anticipated to intensify in the coming weeks.