{"slug": "out-of-sight-still-in-mind-token-compression-for-omni-llms", "title": "Out of Sight, Still in Mind: Token Compression for Omni-LLMs", "summary": "A new research paper targets the input token cost of omni-modal large language models (Omni-LLMs) at inference time, noting that visual tokens account for the vast majority of input and are highly unbalanced against audio and text streams. The work proposes token compression for Omni-LLMs that reason jointly over audio, video and text, aiming to cut the cost of the dominant visual stream.", "body_md": "The goal of this paper is to reduce the input token cost of Omni-modal large language models (Omni-LLMs) at inference time. Omni-LLMs reason jointly over audio, video and text, but the cost of the three streams is highly unbalanced: visual tokens account for the vast majority of the input, and are h", "url": "https://wpnews.pro/news/out-of-sight-still-in-mind-token-compression-for-omni-llms", "canonical_source": "https://aiflash.com/news/154142/", "published_at": "2026-11-19 05:30:02+00:00", "updated_at": "2026-11-19 05:44:42.430578+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-infrastructure", "machine-learning"], "entities": [], "also_reported_by": [], "takeaway": "Omni-LLM input cost is dominated by visual tokens, so compressing the visual stream is the highest-leverage way to cut inference token spend.", "recommendation": {"advice": "Target visual tokens first when trimming omni-LLM inference costs, since the paper attributes the vast majority of input tokens to the visual stream.", "playbook": "llm-cost", "playbook_url": "https://wpnews.pro/playbooks/llm-cost"}, "alternates": {"html": "https://wpnews.pro/news/out-of-sight-still-in-mind-token-compression-for-omni-llms", "markdown": "https://wpnews.pro/news/out-of-sight-still-in-mind-token-compression-for-omni-llms.md", "text": "https://wpnews.pro/news/out-of-sight-still-in-mind-token-compression-for-omni-llms.txt", "jsonld": "https://wpnews.pro/news/out-of-sight-still-in-mind-token-compression-for-omni-llms.jsonld"}}