- Published on
Arena Unveils New Weekly AI Model Rankings: Claude Opus 5.5 Max Tops Agent Leaderboard for First Time, Domestic Model...
- Authors
- Name
- DeepHorus
On October 12, the Arena platform released the AI large model rankings for Week 41 of 2026, covering the period from October 5 to October 11, 2026. In this week's rankings, Anthropic's brand-new model Claude Opus 5.5 Max entered the agent leaderboard for the first time and immediately took first place. The domestic model kimi-k3-max remained firmly in the top 20 of the overall leaderboard, while several domestic models such as qwen3.8-max maintained stable rankings. Overall, domestic models are in the mid-to-upper tier, and the competitive landscape continues to diverge.
The rankings are based on anonymous blind-test voting on the Arena platform, with rankings calculated via ELO; the agent leaderboard uses percentage signals. The rankings reflect users' subjective preferences rather than an absolute capability test. Different sub-leaderboards are independent of one another and should not be compared across leaderboards; score differences within the margin of error are all treated as ties. New models must also accumulate a certain number of votes before they are officially included in the rankings.
At the top of the overall leaderboard, the landscape is relatively stable: gemini-4-argon-high continues to occupy first place with an ELO score of 1525. Several Anthropic models occupy the top four spots; claude-opus-5.5-high rose 2 places to second with an ELO score of 1507, close to third-place claude-opus-4-6-high within the margin of error. This week, gpt-5.5 and gemini-3.5-flash-medium were newly added, ranking 29th and 30th respectively, with ELO scores of 1477 and 1476. Among domestic models, kimi-k3-max ranked 16th with an ELO of 1488; qwen3.8-max was 22nd with an ELO of 1483, up 1 place; Xiaomi's mimo-v2.6-pro was 26th with an ELO of 1480; Zhipu's glm-5.3-max was 27th with an ELO of 1478; all held their rankings steady.
In the coding leaderboard, gemini-4-argon-high continues to sit firmly at the top, with its ELO score rising slightly by 2 points to 1562. Anthropic models showed notable movement: claude-opus-5.5-high surged 7 places into the top five with an ELO score of 1550, only 1 point behind second-place claude-fable-5-high's 1551, treated as a statistical tie. Zhipu's glm-5.3-flash rose 2 places to 27th with an ELO of 1525, becoming the highest-ranked domestic model. This week, claude-opus-4-5-20251101 was newly added, debuting at 30th with an ELO of 1523. Among domestic models, kimi-k3-max was 10th with an ELO of 1540; mimo-v2.6-pro was 11th with an ELO of 1539; deepseek-v4.1-flash-max was 26th with an ELO of 1528; qwen3.8-max was 29th with an ELO of 1524.
In the front-end development leaderboard, the top landscape is stable: claude-opus-5.5-max continues to rank first with an ELO of 1813 and a clear lead. Among domestic models, Xiaomi's mimo-v2.6-flash newly entered the rankings this week, placing 16th with an ELO of 1637. mimo-v2.6-pro rose 5 places to 19th with an ELO of 1629, an impressive performance. Alibaba's qwen3.8-max-0902 rose 1 place to 10th, while qwen3.8-max was 11th; both models entered the top 15, showing that domestic front-end development capabilities continue to improve.
The agent leaderboard saw a major change: Anthropic Claude Opus 5.5 (Max), upon newly entering the rankings, topped the list with a net improvement of 15.53%. Its task confirmation success rate reached 17.29%, and its tool hallucination rate was only 0.37%, with all signal performances at the top level. Claude Opus 5.5 (High) held on to second place with a net improvement of 14.33%, while GPT 6 Astra (Max) rose 1 place to third. Gemini 4 Argon (High) rose 2 places to eighth, with a tool hallucination rate of only 0.18%, making it the model with the lowest tool hallucination rate among current top models. Domestic models are overall in the second tier: Deepseek V4.1 Flash (Max) 17th, Kimi K3 (Max) 18th, Tencent Hy4 preview 19th, Xiaomi MiMo V2.6 Pro 20th, Alibaba Qwen3.8 Max 24th.
At the top of the AI image generation leaderboard, OpenAI models dominate: gpt-image-2.5-sunburst continues to hold first place with an ELO of 1425, leading the second place by more than 20 points. Google's gemini-nano-banana-2.1 newly entered the rankings this week, debuting at sixth with an ELO of 1328. Among domestic models, ByteDance's seedream-5.0-pro ranked 12th and Alibaba's qwen-image-3.0-pro ranked 13th, both in the mid-to-upper range. The top of the AI video leaderboard is still occupied by Google models, with gemini-omni-1.1-flash and gemini-omni-flash ranking first and second respectively, both with ELO scores above 1510. A total of 6 domestic models entered the top 20, among which dreamina-seedance-2.0-720p was fifth, wan3.0 sixth, dreamina-seedance-2.5-720p seventh, minimax-h3 eighth, and happyhorse-1.0 tenth.
Overall, this week's rankings show that Anthropic's new models performed strongly: Claude Opus 5.5 Max topped the agent leaderboard on its debut, and Claude models also occupied top positions in multiple sub-leaderboards. The competitive landscape has tilted further toward Anthropic. Domestic models remained generally stable, with a clear multi-tier distribution; kimi-k3-max, qwen3.8-max, and others continued to hold steady in the mid-to-upper range, and the new model mimo-v2.6-flash also successfully entered the rankings, reflecting the ongoing iteration pace of domestic models. The market is expected to see more domestic large model version updates next week, and it is worth continuing to watch ranking changes.
