Brief, mostly facts
OmniDR-Bench benchmarks agents on video-linked deep research reports
The paper introduces Omni-ART rubric trees for 18 configurations. The best scores 63.3 out of 100, and code and data are on GitHub.
What the research found
Introduces OmniDR-Bench and Omni-ART rubric trees for evaluating agents that produce video-linked deep research reports; best of 18 configurations scores 63.3/100, with code and data released on GitHub.
Claims and how well the evidence supports them
- OmniDR-Bench contains 50 Chinese and English tasks across 19 real-world subcategories 99%
- OmniDR-Bench includes 6,414 criteria 99%
- The best of 18 agent configurations scores 63.3 out of 100 99%
- Code and data are released at github.com/marinero4972/OmniDR-Bench 98%
Evidence
- huggingface.co https://huggingface.co/papers/2610.31158
- github.com https://github.com/marinero4972/OmniDR-Bench
- arxiv.org https://arxiv.org/abs/2610.31158
Actionability rated low.