Dayiheng Liu

3 articles on SOTA Papers

Computer Science

Qwen-CUA Reaches 86.2 on Verified Computer Use

A 397B-A17B mixture-of-experts agent learns screenshot-only keyboard and mouse control from verifiable interactive trajectories, improving OSWorld-Verified performance to 86.2.

cs.LGcs.CL
Aug 8, 20264 min2608.02352
Electrical Engineering

Qwen Audio VAE Compresses Audio Without Bottlenecking Training

A 12.5 Hz continuous latent with encoder pruning cuts 64×30 s encoding latency from 1957 ms to 541 ms.

eess.AS
Jul 28, 20265 min2607.11738
Computer Science

Long-Horizon Workflows Expose Agent Reliability Gaps

OSWorld 2.0 packages 108 multi-application workflows into executable tests, where the best tested agent completes only 20.6%.

cs.AI
Jul 28, 20265 min2606.29537