Peking University and Alibaba unveil OmniDR-Bench for video-centered deep research agents
Qwen · hf · 2026-10-23
Researchers from Peking University and Alibaba TokenHub introduce Video-Centered Deep Research Report Generation, with the OmniDR-Bench benchmark and Omni-ARTs evaluation framework.
- Task: given a video and user request, an agent must identify researchable questions, retrieve external multimodal evidence, and produce an interactive report linked to specific video moments.
- Evaluation: skill-based atomic rubric trees spanning 3 dimensions and 23 shared directions, testing audio-visual grounding, multimodal evidence use, and interactive presentation, built via Audited Rubric Construction (ARC).
- Practical payoff: helps users locate moments, verify claims, and reproduce procedures — filling a gap left by text-only deep research benchmarks.
More from coding & agent
- DeepSeek harness slashes tokens, time and cost vs Codex, but tanks on FrontierCode — teortaxesTex · 2026-10-23
- AI Exposes CVEs Faster Than Maintainers Can Patch: Dev Says AI Does More Harm Than Good — _jaydeepkarale · 2026-10-23
- AI is becoming a mathematical brainstorming companion for algorithm design — Fowe · 2026-10-23
- OpenPrefs MCP centralizes your preferences across Claude, ChatGPT and agents — openprefs · 2026-10-23
- $100K Open Kernel Challenge accused of copying GPU MODE OSS code via AI agent rewrite — josh_wills · 2026-10-23
- GBrain Adds Multi-Account Switching So Your AI Memory Follows Each Identity — garrytan · 2026-10-23