{
  "schema_version": "newruntime-agent-readable-v0.1",
  "type": "raw_signal",
  "id": "tg-2696",
  "slug": "fable-cost-per-completed-task",
  "title": "A pricier model can be cheaper per completed task",
  "description": "Cognition reports that Fable 5 completed coding work with fewer steps and output tokens than its previous lead model.",
  "observed_at": "2026-07-15",
  "why_it_matters": "Agent economics must include turns, tool calls, retries, delegated work, and human review rather than reducing cost to the token price printed on a model card.",
  "novelty": "structural",
  "verification_level": "source-inspected",
  "signal_type": "field-report",
  "evidence_kind": "mixed",
  "status": "published",
  "telegram_message_id": 2696,
  "telegram_url": "https://t.me/qwgai/2696",
  "topics": [
    "agent-economics",
    "coding-agents",
    "model-routing"
  ],
  "entities": [
    "Cognition",
    "Fable 5",
    "Devin Fusion"
  ],
  "related_patterns": [],
  "source_urls": [
    "https://cognition.com/blog/making-fable-cheaper-than-opus"
  ],
  "import_batch": "telegram-2026-07-17-v1",
  "routes": {
    "html": "https://newruntime.com/signals/fable-cost-per-completed-task/",
    "markdown": "https://newruntime.com/signals/fable-cost-per-completed-task.md",
    "json": "https://newruntime.com/signals/fable-cost-per-completed-task.json"
  },
  "source_format": "telegram-export-normalized-json"
}