{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-03-012",
    "source_story_id": "tmp-story-omniseek-audiovisual-tools",
    "edition_id": "mp-2026-10-03-morning-0086",
    "edition_url": "https://themachinepress.com/edition/2026-10-03",
    "position": 12,
    "story_type": "dispatch",
    "section": "frontier-models",
    "editorial_classification": "editorial",
    "headline": "The Model Chose When to Look and When to Listen",
    "slug": "the-model-chose-when-to-look-and-when-to-listen",
    "dek": "OmniSeek retrieves short audio or video windows as tools instead of processing every long recording in one pass.",
    "summary": "OmniSeek retrieves short audio or video windows as tools instead of processing every long recording in one pass.",
    "body_text": "OmniSeek turns evidence acquisition into a multi-turn decision: the model chooses whether to inspect audio or video and which time window to retrieve, then appends the raw segment for further reasoning. Training begins with 170,000 synthesized interleaved trajectories and continues with reinforcement learning that rewards genuinely cross-modal evidence use. The authors report consistent gains across their audio-visual benchmarks, but the result depends on their generated trajectories, tools and evaluation design.",
    "why_it_matters": "OmniSeek retrieves short audio or video windows as tools instead of processing every long recording in one pass.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-03-012/the-model-chose-when-to-look-and-when-to-listen",
    "json_url": "https://themachinepress.com/story/mp-2026-10-03-012.json",
    "first_published_at": "2026-10-03T09:00:00.000-04:00",
    "modified_at": "2026-10-03T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-03-012-001",
        "text": "OmniSeek retrieves short audio or video windows as tools instead of processing every long recording in one pass.",
        "source_ids": [
          "source-2026-10-03-012"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-10-03-012"
    ],
    "tags": [
      "audio visual reasoning",
      "tool use",
      "long context"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-03-012",
      "title": "arXiv preprint 2610.02181",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.02181",
      "canonical_url": "https://arxiv.org/abs/2610.02181",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-01T13:58:16.000-04:00",
      "accessed_at": "2026-10-03T08:31:37.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-03-012-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Model Chose When to Look and When to Listen",
    "publisher": "The Machine Press",
    "published_at": "2026-10-03T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-03-012/the-model-chose-when-to-look-and-when-to-listen"
  }
}