AI
OmniSeek: Native Tool Integration for Multi-turn Audio-Visual Reasoning
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 30/09/2026, 20:00:00 UTC
Briefing created: 02/10/2026, 12:52:26 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/10/2026, 03:18:24 UTC.
Briefings
OmniSeek is an agentic framework that enables the Omni-LLM to perform multi-turn audio-visual reasoning by dynamically selecting which evidence to retrieve and use. The system synthesizes OmniTraj-170K, a corpus of multi-hop Chain-of-Thought trajectories with interleaved audio and visual evidence, to teach the model to seek sparse but critical information across modalities. This process is optimized via two-stage reinforcement learning with verifiable rewards, and an Audio-Visual Necessity objective ensures reasoning depends on both modalities. Experiments show OmniSeek learns adaptive cross-modal evidence seeking and improves audio-visual reasoning performance across various benchmarks.
Key points
- OmniSeek is an agentic framework that transforms an Omni Large Language Model into an active, multi-turn reasoning agent with native tool use.
- The system synthesizes OmniTraj-170K, a corpus of multi-hop Chain-of-Thought trajectories with interleaved audio and visual evidence, to teach the model to seek sparse but critical information across modalities.
Context and limitations
The framework was evaluated across a wide range of benchmarks, but specific performance metrics for each benchmark are not detailed in the provided evidence.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 12:52:26 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeib7zblbfmbl4gek4zb4irtwfwssvz4soijgxivilturicxexzrxjq