SenseTime's new model tested: reads a 582-page prospectus in 8 minutes, but gets "stuck" on making PPTs

What happened: SenseTime released the SenseNova 6.8 Flash multimodal agent model, and zhidx tested it in the SenseTime Xiaohuanxiong desktop app on office tasks. Key numbers: The model read two IPO prospectuses totaling 582 pages in 8 minutes 21 seconds and found 6 updated data points. A PPT task took 64 minutes. Why it matters: SenseTime says the model topped Claude Opus 5, GPT-6 Sol and Kimi K3 on team collaboration and data analysis benchmarks. The test found long PPT formatting took extra time, which may reflect the environment and Harness rather than model speed.

SenseTime released the SenseNova 6.8 Flash multimodal agent model, and zhidx tested it in the SenseTime Xiaohuanxiong desktop app on office tasks.

SenseTime says the model topped Claude Opus 5, GPT-6 Sol and Kimi K3 on team collaboration and data analysis benchmarks. The test found long PPT formatting took extra time, which may reflect the environment and Harness rather than model speed.