{"slug": "mosaic-an-omni-modal-benchmark-for-long-term-memory-and-cross-modal-reasoning-in", "title": "MOSAIC: An Omni-Modal Benchmark for Long-Term Memory and Cross-Modal Reasoning in Conversational Agents", "summary": "Researchers introduced MOSAIC, a benchmark of 923 questions built on persona-grounded conversation histories of 11-24 sessions spanning text, image, video, and audio, designed to diagnose where omni-modal agents lose evidence in long-term memory. MOSAIC uses a matched diagnostic protocol that freezes the question, reader, prompt, and budget while intervening separately on representation, retrieval, and delivery, and its causal validation was strongest for spatial information, with selector-level validation for annotated prosody. The benchmark's decomposition shows representation loss is modality-dependent, residual reading dominates even under gold evidence, and abstention remains weak, so strong answerable scores coexist with failures to refuse.", "body_md": "arXiv:2610.29249v1 Announce Type: new \nAbstract: Omni-modal assistants must remember not only what users say, but also what they show and convey. Existing long-term memory benchmarks are mostly text-and-image or report a single end-to-end score, so they cannot locate whether evidence is lost when it is represented, retrieved, delivered, or read. We introduce MOSAIC, a benchmark of 923 questions over persona-grounded histories of 11-24 sessions spanning text, image, video, and audio. MOSAIC constructs targeted spatial, temporal, and prosodic signals rather than decorative media, with the strongest causal validation for spatial information and selector-level validation for annotated prosody, and pairs them with a matched diagnostic protocol that freezes the question, reader, prompt, and budget while intervening on representation, retrieval, and delivery. Matched decomposition shows that representation loss is modality-dependent, residual reading dominates even under gold evidence, and abstention remains weak on these controls: strong answerable scores coexist with failures to refuse. MOSAIC thus turns omni-modal memory evaluation from a leaderboard into a diagnosis of write, retrieve, deliver, and read failures.", "url": "https://wpnews.pro/news/mosaic-an-omni-modal-benchmark-for-long-term-memory-and-cross-modal-reasoning-in", "canonical_source": "https://www.machinebrief.com/news/mosaic-an-omni-modal-benchmark-for-long-term-memory-and-cros-ohch", "published_at": "2026-10-22 04:00:00+00:00", "updated_at": "2026-10-22 04:41:31.633996+00:00", "lang": "en", "topics": ["ai-research", "large-language-models", "ai-agents", "ai-safety"], "entities": ["MOSAIC"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/mosaic-an-omni-modal-benchmark-for-long-term-memory-and-cross-modal-reasoning-in", "markdown": "https://wpnews.pro/news/mosaic-an-omni-modal-benchmark-for-long-term-memory-and-cross-modal-reasoning-in.md", "text": "https://wpnews.pro/news/mosaic-an-omni-modal-benchmark-for-long-term-memory-and-cross-modal-reasoning-in.txt", "jsonld": "https://wpnews.pro/news/mosaic-an-omni-modal-benchmark-for-long-term-memory-and-cross-modal-reasoning-in.jsonld"}}