ByteDance's Doubao Work secured first place in the October 2026 SuperCLUE-Office end-to-end office-agent benchmark, achieving an overall score of 85.33. According to results published on the SuperCLUE site and reported by IT Home, Moonshot AI's Kimi Work followed in second place with a score of 84.57, while Tencent's WorkBuddy took third with 83.99.
The SuperCLUE-Office benchmark evaluates the ability of office agents to complete actual deliverables in formats such as Word, PowerPoint, Excel, and charts. The October test cycle assessed three dimensions—document processing, materials drafting, and data analysis—across 13 secondary tasks. The evaluation scored five specific capability areas: file and environment operations, multi-file planning, constraint following, content quality, and safety and permission boundaries.
Eight Chinese office agents participated in the benchmark: Doubao Work, Kimi Work, WorkBuddy, Baidu Dazi, Alibaba's QwenWork, SenseTime's Raccoon Work, iFLYTEK Spark Work, and 360's Nano Work. While the top four agents finished within 2.65 points of each other, SuperCLUE noted a clear gradient below the leaders, with a gap of 32.20 points between the highest and lowest overall scores.
The remaining scores were as follows: Baidu Dazi scored 82.68, QwenWork scored 78.38, Raccoon Work scored 69.20, iFLYTEK Spark Work scored 69.12, and Nano Work scored 53.13.
Performance varied by category among the leaders. Doubao Work led in document processing with a score of 87.12, nearly five points ahead of the next entry, though Baidu Dazi and WorkBuddy remained close behind. In that dimension, all eight products scored above 72. Kimi Work led in data analysis with 88.00, followed by WorkBuddy at 86.13 and Doubao Work at 81.50. SuperCLUE reported that data analysis had the widest spread, featuring a 72-point gap between the best and worst scores and an average of 71.72. Materials drafting showed a tighter top end but an uneven distribution further down, with a 47.75-point maximum gap and a 73.62 average.
SuperCLUE stated that office agents are transitioning into everyday workflows but remain uneven when performing harder analysis and drafting jobs, which the group flagged as the next competitive focus. Pandaily noted that because these results originate from a single organizer's harness and task set, they should be viewed as one public leaderboard rather than a universal ranking of every office agent.
