{
  "schema_version": "1.1",
  "id": "archive:https://www.youtube.com/watch?v=QHBjufYK8TA",
  "slug": "the-state-of-model-routing-nvidia-cognition-openrouter-0eyef50",
  "url": "https://feed7.dev/p/the-state-of-model-routing-nvidia-cognition-openrouter-0eyef50",
  "title": "The State of Model Routing — NVIDIA, Cognition, OpenRouter",
  "why_included": "Model routing can cut agent costs only when delegation, escalation, and context transfer are designed together; duplicating context across models can erase the savings.",
  "summary": "The panel describes routing work between small models and frontier models, with compact handoffs instead of copying full traces. It notes that provider KV caches commonly have a **5-minute lifetime**, though that is an operational policy rather than a physical limit.",
  "practical_implication": "Give most context to one model, let workers return file references or compact findings, and keep a frontier model available for escalation. Measure total task cost: larger models may use tools and tokens more efficiently even when their per-token price is higher.",
  "agent_context": "The panel describes routing work between small models and frontier models, with compact handoffs instead of copying full traces. It notes that provider KV caches commonly have a **5-minute lifetime**, though that is an operational policy rather than a physical limit.\n\nGive most context to one model, let workers return file references or compact findings, and keep a frontier model available for escalation. Measure total task cost: larger models may use tools and tokens more efficiently even when their per-token price is higher.\n\nRouting adds a controller to an already nondeterministic system. Small models may not recognize when they are outside their competence, repeated file reads can multiply cost, and the panel characterizes current orchestration techniques as early.",
  "source": {
    "name": "AI Engineer",
    "url": "https://www.youtube.com/watch?v=QHBjufYK8TA",
    "published_at": "2026-08-06T17:07:24.000Z"
  },
  "source_class": "video",
  "content_type": "Video",
  "layer": "agent",
  "domains": [
    "coding"
  ],
  "topics": [
    "multi-agent",
    "model-selection",
    "context-caching"
  ],
  "verification": {
    "status": "source_linked",
    "label": "Source Linked",
    "method": "source_feed",
    "verified_at": null
  },
  "uncertainty": [
    "Routing adds a controller to an already nondeterministic system. Small models may not recognize when they are outside their competence, repeated file reads can multiply cost, and the panel characterizes current orchestration techniques as early."
  ],
  "connected_context": {
    "meaning": "This makes routing a context-management and total-task economics problem, not simply a per-token price choice. It reinforces planner–worker separation with compact handoffs and frontier escalation, while narrowing confidence in routing gains: controller nondeterminism, repeated reads, short-lived provider caches, and weak self-recognition by small models can erase savings.",
    "corpus_size": 409,
    "generated_at": "2026-08-10T10:05:29.932Z",
    "connections": [
      {
        "title": "Agent swarms and the new model economics",
        "source_name": "Cursor",
        "source_url": "https://cursor.com/blog/agent-swarm-model-economics",
        "feed7_url": "https://feed7.dev/p/agent-swarm-model-economics-02n8muc",
        "reason": "Cursor’s planner–worker context separation reinforces the panel’s recommendation to concentrate context in one model and return compact worker results, while its findings support evaluating harness costs alongside model prices."
      },
      {
        "title": "How Cursor Router chooses the right model for the task",
        "source_name": "Cursor",
        "source_url": "https://cursor.com/blog/how-cursor-router-works",
        "feed7_url": "https://feed7.dev/p/how-cursor-router-works-1fhejk3",
        "reason": "Cursor supplies the production correction signals, cache accounting, and task-level performance learning needed to address the panel’s warning that static routing and small-model self-assessment are unreliable."
      },
      {
        "title": "Multiplayer agentic engineering — Arjun Singh, Superconductor",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=OL7kfezynJM",
        "feed7_url": "https://feed7.dev/p/multiplayer-agentic-engineering-arjun-singh-superconductor-01w4k91",
        "reason": "Superconductor’s codebase-specific benchmarking provides an implementation path for the panel’s call to route by demonstrated task fit rather than generic model assumptions."
      },
      {
        "title": "Agora: Enhancing LLM Agent Reasoning Via Auction-Based Task Allocation",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.09600v1",
        "feed7_url": "https://feed7.dev/p/2607-09600v1-0uoqpsx",
        "reason": "Agora proposes a more formal cost–quality controller for expert allocation, but the panel’s operational cautions show that benchmark routing gains do not by themselves resolve added-controller nondeterminism or context-transfer costs."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-08-06T17:07:24.000Z",
  "modified_at": "2026-08-06T17:07:24.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/the-state-of-model-routing-nvidia-cognition-openrouter-0eyef50",
    "json": "https://feed7.dev/p/the-state-of-model-routing-nvidia-cognition-openrouter-0eyef50.json",
    "markdown": "https://feed7.dev/p/the-state-of-model-routing-nvidia-cognition-openrouter-0eyef50.md"
  }
}