{
 "Long Agentic": {
  "parts": [
   {
    "name": "Finance & Compliance Apps",
    "gloss": "Payroll, budgets, tax claims and audit records: keep the money and the paper trail right as the rules change, turn after turn.",
    "tasks": 3
   },
   {
    "name": "Operations & Commerce Apps",
    "gloss": "Ticketing, equipment hire, stock and site records: keep a live operational app correct as the business changes around it.",
    "tasks": 4
   },
   {
    "name": "Long-Run Operations",
    "gloss": "Run an existing system through simulated months of releases, closes and change requests, where early choices compound.",
    "tasks": 3
   },
   {
    "name": "Delegated Operations",
    "gloss": "Keep a small business's books and duties right for weeks while its owner is away, through a command someone else follows to the letter; what matters shows only in the data.",
    "tasks": 4
   },
   {
    "name": "Systems Engineering",
    "gloss": "Build and repair complex systems from a specification, then keep them running under realistic load and failures.",
    "tasks": 2
   },
   {
    "name": "Performance Optimisation",
    "gloss": "Make real programs faster and real schedules tighter, scored on measured results rather than judgement.",
    "tasks": 2
   }
  ],
  "models": {
   "claude-fable-5.1": {
    "scores": {
     "Finance & Compliance Apps": 83.3,
     "Operations & Commerce Apps": 87.5,
     "Long-Run Operations": 54.6,
     "Delegated Operations": 79.0,
     "Systems Engineering": 87.6,
     "Performance Optimisation": 73.7
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Finance & Compliance Apps": 38.9,
     "Operations & Commerce Apps": 50.0,
     "Long-Run Operations": 36.9,
     "Delegated Operations": 7.5,
     "Systems Engineering": 74.4,
     "Performance Optimisation": 56.6
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Finance & Compliance Apps": 100.0,
     "Operations & Commerce Apps": 87.5,
     "Long-Run Operations": 53.8,
     "Delegated Operations": 34.2,
     "Systems Engineering": 97.1,
     "Performance Optimisation": 83.6
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Finance & Compliance Apps": 83.3,
     "Operations & Commerce Apps": 79.2,
     "Long-Run Operations": 48.9,
     "Delegated Operations": 30.9,
     "Systems Engineering": 77.3,
     "Performance Optimisation": 74.6
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Finance & Compliance Apps": 27.8,
     "Operations & Commerce Apps": 43.8,
     "Long-Run Operations": 38.3,
     "Delegated Operations": 25.6,
     "Systems Engineering": 72.4,
     "Performance Optimisation": 57.4
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "Finance & Compliance Apps": 5.6,
     "Operations & Commerce Apps": 4.2,
     "Long-Run Operations": 14.5,
     "Delegated Operations": 1.0,
     "Systems Engineering": 68.7,
     "Performance Optimisation": 39.8
    }
   },
   "gemini-4-argon": {
    "scores": {
     "Finance & Compliance Apps": 88.9,
     "Operations & Commerce Apps": 70.8,
     "Long-Run Operations": 27.1,
     "Delegated Operations": 46.9,
     "Systems Engineering": 96.1,
     "Performance Optimisation": 45.4
    }
   },
   "gemma-4-31b": {
    "scores": {
     "Finance & Compliance Apps": 5.6,
     "Operations & Commerce Apps": 8.3,
     "Long-Run Operations": 0.7,
     "Delegated Operations": 4.8,
     "Systems Engineering": 28.3,
     "Performance Optimisation": 32.3
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 20.9,
     "Long-Run Operations": 5.4,
     "Delegated Operations": 2.8,
     "Systems Engineering": 75.3,
     "Performance Optimisation": 40.1
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "Finance & Compliance Apps": 27.8,
     "Operations & Commerce Apps": 33.3,
     "Long-Run Operations": 16.9,
     "Delegated Operations": 22.9,
     "Systems Engineering": 72.3,
     "Performance Optimisation": 45.7
    }
   },
   "gpt-6-astra": {
    "scores": {
     "Finance & Compliance Apps": 100.0,
     "Operations & Commerce Apps": 100.0,
     "Long-Run Operations": 33.4,
     "Delegated Operations": 68.9,
     "Systems Engineering": 97.4,
     "Performance Optimisation": 61.8
    }
   },
   "gpt-6-luna": {
    "scores": {
     "Finance & Compliance Apps": 38.9,
     "Operations & Commerce Apps": 43.8,
     "Long-Run Operations": 17.6,
     "Delegated Operations": 0.8,
     "Systems Engineering": 60.1,
     "Performance Optimisation": 43.9
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Finance & Compliance Apps": 66.7,
     "Operations & Commerce Apps": 100.0,
     "Long-Run Operations": 25.0,
     "Delegated Operations": 56.2,
     "Systems Engineering": 96.6,
     "Performance Optimisation": 71.7
    }
   },
   "gpt-6.1-terra": {
    "scores": {
     "Finance & Compliance Apps": 55.6,
     "Operations & Commerce Apps": 81.2,
     "Long-Run Operations": 28.2,
     "Delegated Operations": 0.8,
     "Systems Engineering": 97.2,
     "Performance Optimisation": 50.6
    }
   },
   "grok-4.6": {
    "scores": {
     "Finance & Compliance Apps": 44.4,
     "Operations & Commerce Apps": 18.8,
     "Long-Run Operations": 11.6,
     "Delegated Operations": 4.4,
     "Systems Engineering": 71.2,
     "Performance Optimisation": 50.6
    }
   },
   "grok-4.7": {
    "scores": {
     "Finance & Compliance Apps": 61.1,
     "Operations & Commerce Apps": 50.0,
     "Long-Run Operations": 9.3,
     "Delegated Operations": 48.4,
     "Systems Engineering": 70.1,
     "Performance Optimisation": 42.3
    }
   },
   "inkling": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 1.9,
     "Delegated Operations": 0.6,
     "Systems Engineering": 38.8,
     "Performance Optimisation": 34.1
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Finance & Compliance Apps": 44.4,
     "Operations & Commerce Apps": 27.1,
     "Long-Run Operations": 27.3,
     "Delegated Operations": 29.5,
     "Systems Engineering": 74.9,
     "Performance Optimisation": 56.1
    }
   },
   "kimi-k3": {
    "scores": {
     "Finance & Compliance Apps": 22.2,
     "Operations & Commerce Apps": 16.7,
     "Long-Run Operations": 13.2,
     "Delegated Operations": 3.5,
     "Systems Engineering": 82.7,
     "Performance Optimisation": 61.1
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "Finance & Compliance Apps": 33.3,
     "Operations & Commerce Apps": 50.0,
     "Long-Run Operations": 18.3,
     "Delegated Operations": 8.3,
     "Systems Engineering": 71.9,
     "Performance Optimisation": 50.6
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "Finance & Compliance Apps": 5.6,
     "Operations & Commerce Apps": 8.3,
     "Long-Run Operations": 5.8,
     "Delegated Operations": 51.0,
     "Systems Engineering": 67.1,
     "Performance Optimisation": 56.5
    }
   },
   "mistral-large-4": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.9,
     "Delegated Operations": 4.9,
     "Systems Engineering": 29.4,
     "Performance Optimisation": 38.3
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "Finance & Compliance Apps": 22.2,
     "Operations & Commerce Apps": 29.2,
     "Long-Run Operations": 9.8,
     "Delegated Operations": 14.9,
     "Systems Engineering": 58.0,
     "Performance Optimisation": 35.7
    }
   },
   "muse-spark-open": {
    "scores": {
     "Finance & Compliance Apps": 11.1,
     "Operations & Commerce Apps": 4.2,
     "Long-Run Operations": 2.1,
     "Delegated Operations": 0.6,
     "Systems Engineering": 38.6,
     "Performance Optimisation": 37.4
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "Finance & Compliance Apps": 72.2,
     "Operations & Commerce Apps": 52.1,
     "Long-Run Operations": 26.7,
     "Delegated Operations": 46.8,
     "Systems Engineering": 86.4,
     "Performance Optimisation": 41.2
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 1.1,
     "Delegated Operations": 4.8,
     "Systems Engineering": 23.8,
     "Performance Optimisation": 30.4
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 6.2,
     "Long-Run Operations": 5.3,
     "Delegated Operations": 5.7,
     "Systems Engineering": 36.9,
     "Performance Optimisation": 36.2
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Finance & Compliance Apps": 5.6,
     "Operations & Commerce Apps": 31.2,
     "Long-Run Operations": 3.6,
     "Delegated Operations": 1.6,
     "Systems Engineering": 64.0,
     "Performance Optimisation": 41.8
    }
   },
   "qwen3.8-max": {
    "scores": {
     "Finance & Compliance Apps": 27.8,
     "Operations & Commerce Apps": 33.3,
     "Long-Run Operations": 18.9,
     "Delegated Operations": 7.2,
     "Systems Engineering": 74.4,
     "Performance Optimisation": 43.7
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.8,
     "Delegated Operations": 5.6,
     "Systems Engineering": 27.0,
     "Performance Optimisation": 36.1
    }
   },
   "step-5-preview": {
    "scores": {
     "Finance & Compliance Apps": 5.6,
     "Operations & Commerce Apps": 20.8,
     "Long-Run Operations": 9.9,
     "Delegated Operations": 1.4,
     "Systems Engineering": 63.5,
     "Performance Optimisation": 43.0
    }
   }
  }
 },
 "Reasoning": {
  "parts": [
   {
    "name": "Theory of mind",
    "gloss": "Track who knows or believes what in a short story.",
    "tasks": 52
   },
   {
    "name": "Zebra puzzles",
    "gloss": "Deduce who has which attribute from interlocking clues.",
    "tasks": 50
   },
   {
    "name": "Spatial",
    "gloss": "Reason about shapes and positions in 2D and 3D.",
    "tasks": 50
   },
   {
    "name": "Logic with navigation",
    "gloss": "Work out who tells the truth from a chain of statements.",
    "tasks": 50
   }
  ],
  "models": {
   "claude-fable-5.1": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 86.0
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Theory of mind": 78.8,
     "Zebra puzzles": 96.8,
     "Spatial": 88.0,
     "Logic with navigation": 66.0
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 86.0
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 82.0
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Theory of mind": 88.5,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 70.0
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "Theory of mind": 76.9,
     "Zebra puzzles": 98.2,
     "Spatial": 100.0,
     "Logic with navigation": 82.0
    }
   },
   "gemini-4-argon": {
    "scores": {
     "Theory of mind": 92.3,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 88.0
    }
   },
   "gemma-4-31b": {
    "scores": {
     "Theory of mind": 73.1,
     "Zebra puzzles": 36.0,
     "Spatial": 96.0,
     "Logic with navigation": 68.0
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "Theory of mind": 73.1,
     "Zebra puzzles": 79.5,
     "Spatial": 96.0,
     "Logic with navigation": 62.0
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 76.0
    }
   },
   "gpt-6-astra": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 88.0
    }
   },
   "gpt-6-luna": {
    "scores": {
     "Theory of mind": 73.1,
     "Zebra puzzles": 96.0,
     "Spatial": 96.0,
     "Logic with navigation": 62.0
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 86.0
    }
   },
   "gpt-6.1-terra": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 72.0
    }
   },
   "grok-4.6": {
    "scores": {
     "Theory of mind": 90.4,
     "Zebra puzzles": 96.5,
     "Spatial": 98.0,
     "Logic with navigation": 80.0
    }
   },
   "grok-4.7": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 86.0,
     "Spatial": 94.0,
     "Logic with navigation": 66.0
    }
   },
   "inkling": {
    "scores": {
     "Theory of mind": 65.4,
     "Zebra puzzles": 88.0,
     "Spatial": 94.0,
     "Logic with navigation": 66.0
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 78.0
    }
   },
   "kimi-k3": {
    "scores": {
     "Theory of mind": 82.7,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 80.0
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 72.0,
     "Spatial": 94.0,
     "Logic with navigation": 66.0
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "Theory of mind": 63.5,
     "Zebra puzzles": 100.0,
     "Spatial": 96.0,
     "Logic with navigation": 72.0
    }
   },
   "mistral-large-4": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 93.0,
     "Spatial": 94.0,
     "Logic with navigation": 68.0
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 98.0,
     "Spatial": 100.0,
     "Logic with navigation": 76.0
    }
   },
   "muse-spark-open": {
    "scores": {
     "Theory of mind": 76.9,
     "Zebra puzzles": 40.0,
     "Spatial": 100.0,
     "Logic with navigation": 78.0
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 100.0,
     "Spatial": 96.0,
     "Logic with navigation": 80.0
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "Theory of mind": 63.5,
     "Zebra puzzles": 63.2,
     "Spatial": 100.0,
     "Logic with navigation": 60.0
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "Theory of mind": 75.0,
     "Zebra puzzles": 67.5,
     "Spatial": 98.0,
     "Logic with navigation": 58.0
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Theory of mind": 59.6,
     "Zebra puzzles": 96.5,
     "Spatial": 96.0,
     "Logic with navigation": 68.0
    }
   },
   "qwen3.8-max": {
    "scores": {
     "Theory of mind": 78.8,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 74.0
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "Theory of mind": 75.0,
     "Zebra puzzles": 67.5,
     "Spatial": 98.0,
     "Logic with navigation": 60.0
    }
   },
   "step-5-preview": {
    "scores": {
     "Theory of mind": 75.0,
     "Zebra puzzles": 90.0,
     "Spatial": 98.0,
     "Logic with navigation": 72.0
    }
   }
  }
 },
 "Coding": {
  "parts": [
   {
    "name": "LLM API Integration",
    "gloss": "Provider wire formats and request/response handling.",
    "tasks": 4
   },
   {
    "name": "Agent Loop",
    "gloss": "Agent runtime control flow.",
    "tasks": 5
   },
   {
    "name": "Billing & Limits",
    "gloss": "Metering, quotas and limit enforcement.",
    "tasks": 4
   },
   {
    "name": "Cloud Infrastructure",
    "gloss": "Storage, compute and cloud service plumbing.",
    "tasks": 6
   },
   {
    "name": "Product & UX",
    "gloss": "User-facing product behaviour.",
    "tasks": 6
   },
   {
    "name": "Memory & Tools",
    "gloss": "Memory handling and tool integrations.",
    "tasks": 4
   }
  ],
  "models": {
   "claude-fable-5.1": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 83.3,
     "Memory & Tools": 100.0
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 95.8,
     "Cloud Infrastructure": 83.3,
     "Product & UX": 77.8,
     "Memory & Tools": 100.0
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "LLM API Integration": 97.5,
     "Agent Loop": 96.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 88.9,
     "Memory & Tools": 97.5
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 83.3,
     "Cloud Infrastructure": 98.3,
     "Product & UX": 87.2,
     "Memory & Tools": 100.0
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "LLM API Integration": 87.5,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 88.9,
     "Memory & Tools": 100.0
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "LLM API Integration": 87.5,
     "Agent Loop": 80.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 72.2,
     "Memory & Tools": 75.0
    }
   },
   "gemini-4-argon": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 80.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 77.8,
     "Memory & Tools": 83.3
    }
   },
   "gemma-4-31b": {
    "scores": {
     "LLM API Integration": 54.2,
     "Agent Loop": 50.0,
     "Billing & Limits": 25.0,
     "Cloud Infrastructure": 63.9,
     "Product & UX": 33.3,
     "Memory & Tools": 67.9
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 80.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 83.3,
     "Product & UX": 50.0,
     "Memory & Tools": 75.0
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "LLM API Integration": 87.5,
     "Agent Loop": 100.0,
     "Billing & Limits": 79.2,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 72.2,
     "Memory & Tools": 75.0
    }
   },
   "gpt-6-astra": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 90.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 94.5,
     "Memory & Tools": 100.0
    }
   },
   "gpt-6-luna": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 90.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 83.3,
     "Memory & Tools": 50.0
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "LLM API Integration": 83.3,
     "Agent Loop": 90.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 100.0,
     "Memory & Tools": 100.0
    }
   },
   "gpt-6.1-terra": {
    "scores": {
     "LLM API Integration": 83.3,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 83.3,
     "Memory & Tools": 100.0
    }
   },
   "grok-4.6": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 80.0,
     "Billing & Limits": 95.8,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 83.3,
     "Memory & Tools": 100.0
    }
   },
   "grok-4.7": {
    "scores": {
     "LLM API Integration": 37.5,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 66.7,
     "Memory & Tools": 100.0
    }
   },
   "inkling": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 60.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 66.7,
     "Product & UX": 47.2,
     "Memory & Tools": 67.9
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 80.0,
     "Billing & Limits": 95.8,
     "Cloud Infrastructure": 92.8,
     "Product & UX": 55.6,
     "Memory & Tools": 100.0
    }
   },
   "kimi-k3": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 40.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 50.0,
     "Product & UX": 50.0,
     "Memory & Tools": 25.0
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 90.0,
     "Product & UX": 72.2,
     "Memory & Tools": 72.5
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 98.3,
     "Product & UX": 94.4,
     "Memory & Tools": 68.9
    }
   },
   "mistral-large-4": {
    "scores": {
     "LLM API Integration": 68.3,
     "Agent Loop": 56.0,
     "Billing & Limits": 45.8,
     "Cloud Infrastructure": 47.2,
     "Product & UX": 55.6,
     "Memory & Tools": 96.4
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 95.8,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 72.2,
     "Memory & Tools": 100.0
    }
   },
   "muse-spark-open": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 78.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 91.7,
     "Product & UX": 72.2,
     "Memory & Tools": 90.2
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "LLM API Integration": 62.5,
     "Agent Loop": 80.0,
     "Billing & Limits": 95.8,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 94.4,
     "Memory & Tools": 75.0
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "LLM API Integration": 66.7,
     "Agent Loop": 40.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 91.7,
     "Product & UX": 33.3,
     "Memory & Tools": 90.2
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "LLM API Integration": 66.7,
     "Agent Loop": 70.0,
     "Billing & Limits": 70.8,
     "Cloud Infrastructure": 63.9,
     "Product & UX": 66.7,
     "Memory & Tools": 96.4
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "LLM API Integration": 87.5,
     "Agent Loop": 100.0,
     "Billing & Limits": 79.2,
     "Cloud Infrastructure": 86.1,
     "Product & UX": 72.2,
     "Memory & Tools": 100.0
    }
   },
   "qwen3.8-max": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 60.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 33.3,
     "Product & UX": 50.0,
     "Memory & Tools": 75.0
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 50.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 80.6,
     "Product & UX": 83.3,
     "Memory & Tools": 90.2
    }
   },
   "step-5-preview": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 70.8,
     "Cloud Infrastructure": 77.8,
     "Product & UX": 72.2,
     "Memory & Tools": 100.0
    }
   }
  }
 },
 "Browser": {
  "parts": [
   {
    "name": "Data Aggregation",
    "gloss": "Exhaustive traversal of a large record set and multi-part roll-ups.",
    "tasks": 6
   },
   {
    "name": "Sequential Planning",
    "gloss": "Order-dependent steps where each step constrains the next.",
    "tasks": 4
   },
   {
    "name": "Analysis-Driven Action",
    "gloss": "Derive a result from the data, then act on it in the app.",
    "tasks": 7
   },
   {
    "name": "Structural Integrity",
    "gloss": "Keep a document or record graph consistent across an edit.",
    "tasks": 5
   },
   {
    "name": "Constraint Adherence",
    "gloss": "Exact fidelity, plus correctly declining what the app won't permit.",
    "tasks": 3
   }
  ],
  "models": {
   "claude-fable-5.1": {
    "scores": {
     "Data Aggregation": 83.3,
     "Sequential Planning": 71.4,
     "Analysis-Driven Action": 97.8,
     "Structural Integrity": 94.8,
     "Constraint Adherence": 100.0
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Data Aggregation": 40.5,
     "Sequential Planning": 57.4,
     "Analysis-Driven Action": 88.8,
     "Structural Integrity": 91.7,
     "Constraint Adherence": 61.8
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Data Aggregation": 99.2,
     "Sequential Planning": 69.6,
     "Analysis-Driven Action": 80.7,
     "Structural Integrity": 95.6,
     "Constraint Adherence": 97.9
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Data Aggregation": 82.6,
     "Sequential Planning": 69.6,
     "Analysis-Driven Action": 96.4,
     "Structural Integrity": 97.3,
     "Constraint Adherence": 99.3
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Data Aggregation": 66.7,
     "Sequential Planning": 89.3,
     "Analysis-Driven Action": 56.5,
     "Structural Integrity": 52.1,
     "Constraint Adherence": 88.9
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "Data Aggregation": 39.4,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 47.5,
     "Structural Integrity": 53.8,
     "Constraint Adherence": 96.5
    }
   },
   "gemini-4-argon": {
    "scores": {
     "Data Aggregation": 83.3,
     "Sequential Planning": 83.9,
     "Analysis-Driven Action": 74.5,
     "Structural Integrity": 86.1,
     "Constraint Adherence": 93.8
    }
   },
   "gemma-4-31b": {
    "scores": {
     "Data Aggregation": 0.0,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "Data Aggregation": 48.9,
     "Sequential Planning": 54.6,
     "Analysis-Driven Action": 65.5,
     "Structural Integrity": 48.9,
     "Constraint Adherence": 93.8
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "Data Aggregation": 56.1,
     "Sequential Planning": 78.7,
     "Analysis-Driven Action": 76.9,
     "Structural Integrity": 57.0,
     "Constraint Adherence": 94.4
    }
   },
   "gpt-6-astra": {
    "scores": {
     "Data Aggregation": 83.3,
     "Sequential Planning": 96.4,
     "Analysis-Driven Action": 97.1,
     "Structural Integrity": 83.6,
     "Constraint Adherence": 100.0
    }
   },
   "gpt-6-luna": {
    "scores": {
     "Data Aggregation": 72.8,
     "Sequential Planning": 83.2,
     "Analysis-Driven Action": 58.4,
     "Structural Integrity": 57.3,
     "Constraint Adherence": 97.9
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Data Aggregation": 83.3,
     "Sequential Planning": 96.4,
     "Analysis-Driven Action": 97.1,
     "Structural Integrity": 93.1,
     "Constraint Adherence": 100.0
    }
   },
   "gpt-6.1-terra": {
    "scores": {
     "Data Aggregation": 91.3,
     "Sequential Planning": 93.8,
     "Analysis-Driven Action": 90.4,
     "Structural Integrity": 94.6,
     "Constraint Adherence": 100.0
    }
   },
   "grok-4.6": {
    "scores": {
     "Data Aggregation": 17.6,
     "Sequential Planning": 14.8,
     "Analysis-Driven Action": 59.3,
     "Structural Integrity": 72.9,
     "Constraint Adherence": 96.5
    }
   },
   "grok-4.7": {
    "scores": {
     "Data Aggregation": 68.2,
     "Sequential Planning": 17.5,
     "Analysis-Driven Action": 6.3,
     "Structural Integrity": 97.3,
     "Constraint Adherence": 74.7
    }
   },
   "inkling": {
    "scores": {
     "Data Aggregation": 50.0,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Data Aggregation": 83.3,
     "Sequential Planning": 91.3,
     "Analysis-Driven Action": 63.2,
     "Structural Integrity": 69.0,
     "Constraint Adherence": 81.9
    }
   },
   "kimi-k3": {
    "scores": {
     "Data Aggregation": 66.7,
     "Sequential Planning": 59.2,
     "Analysis-Driven Action": 88.6,
     "Structural Integrity": 75.5,
     "Constraint Adherence": 81.6
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "Data Aggregation": 57.4,
     "Sequential Planning": 67.1,
     "Analysis-Driven Action": 81.1,
     "Structural Integrity": 94.8,
     "Constraint Adherence": 93.1
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "Data Aggregation": 82.6,
     "Sequential Planning": 69.4,
     "Analysis-Driven Action": 80.7,
     "Structural Integrity": 76.5,
     "Constraint Adherence": 98.6
    }
   },
   "mistral-large-4": {
    "scores": {
     "Data Aggregation": 0.0,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 3.0,
     "Constraint Adherence": 0.0
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "Data Aggregation": 35.2,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 6.2,
     "Structural Integrity": 33.6,
     "Constraint Adherence": 83.3
    }
   },
   "muse-spark-open": {
    "scores": {
     "Data Aggregation": 34.7,
     "Sequential Planning": 14.8,
     "Analysis-Driven Action": 17.6,
     "Structural Integrity": 21.5,
     "Constraint Adherence": 52.1
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "Data Aggregation": 22.0,
     "Sequential Planning": 71.4,
     "Analysis-Driven Action": 58.9,
     "Structural Integrity": 75.5,
     "Constraint Adherence": 97.9
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "Data Aggregation": 0.7,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 15.8,
     "Structural Integrity": 13.7,
     "Constraint Adherence": 89.1
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "Data Aggregation": 17.8,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 24.3,
     "Structural Integrity": 21.7,
     "Constraint Adherence": 51.4
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Data Aggregation": 33.3,
     "Sequential Planning": 7.4,
     "Analysis-Driven Action": 15.7,
     "Structural Integrity": 36.7,
     "Constraint Adherence": 27.8
    }
   },
   "qwen3.8-max": {
    "scores": {
     "Data Aggregation": 39.9,
     "Sequential Planning": 31.6,
     "Analysis-Driven Action": 59.0,
     "Structural Integrity": 45.3,
     "Constraint Adherence": 76.0
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "Data Aggregation": 1.1,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 35.8,
     "Structural Integrity": 32.1,
     "Constraint Adherence": 74.1
    }
   },
   "step-5-preview": {
    "scores": {
     "Data Aggregation": 18.0,
     "Sequential Planning": 3.4,
     "Analysis-Driven Action": 17.9,
     "Structural Integrity": 45.2,
     "Constraint Adherence": 49.3
    }
   }
  }
 },
 "Data Analysis": {
  "parts": [
   {
    "name": "Consecutive events",
    "gloss": "Find users or entities with runs of consecutive growth in a small, messy table of dated values.",
    "tasks": 47
   },
   {
    "name": "Table join",
    "gloss": "Map the columns of one table onto the obfuscated columns of another.",
    "tasks": 52
   },
   {
    "name": "Table reformat",
    "gloss": "Convert a table between CSV, TSV, JSON, JSONL, Markdown and HTML.",
    "tasks": 51
   }
  ],
  "models": {
   "claude-fable-5.1": {
    "scores": {
     "Consecutive events": 90.6,
     "Table join": 56.2,
     "Table reformat": 94.1
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Consecutive events": 16.4,
     "Table join": 44.8,
     "Table reformat": 98.0
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Consecutive events": 90.6,
     "Table join": 52.3,
     "Table reformat": 98.0
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Consecutive events": 40.6,
     "Table join": 39.8,
     "Table reformat": 98.0
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Consecutive events": 90.3,
     "Table join": 48.2,
     "Table reformat": 100.0
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "Consecutive events": 20.6,
     "Table join": 45.3,
     "Table reformat": 96.1
    }
   },
   "gemini-4-argon": {
    "scores": {
     "Consecutive events": 91.4,
     "Table join": 69.9,
     "Table reformat": 94.1
    }
   },
   "gemma-4-31b": {
    "scores": {
     "Consecutive events": 13.6,
     "Table join": 46.3,
     "Table reformat": 100.0
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "Consecutive events": 84.3,
     "Table join": 48.8,
     "Table reformat": 96.1
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "Consecutive events": 87.7,
     "Table join": 50.2,
     "Table reformat": 100.0
    }
   },
   "gpt-6-astra": {
    "scores": {
     "Consecutive events": 90.6,
     "Table join": 58.4,
     "Table reformat": 100.0
    }
   },
   "gpt-6-luna": {
    "scores": {
     "Consecutive events": 70.5,
     "Table join": 49.6,
     "Table reformat": 100.0
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Consecutive events": 90.6,
     "Table join": 57.4,
     "Table reformat": 100.0
    }
   },
   "gpt-6.1-terra": {
    "scores": {
     "Consecutive events": 90.5,
     "Table join": 56.5,
     "Table reformat": 100.0
    }
   },
   "grok-4.6": {
    "scores": {
     "Consecutive events": 84.6,
     "Table join": 43.1,
     "Table reformat": 98.0
    }
   },
   "grok-4.7": {
    "scores": {
     "Consecutive events": 85.7,
     "Table join": 44.9,
     "Table reformat": 100.0
    }
   },
   "inkling": {
    "scores": {
     "Consecutive events": 69.9,
     "Table join": 48.5,
     "Table reformat": 100.0
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Consecutive events": 90.2,
     "Table join": 49.1,
     "Table reformat": 98.0
    }
   },
   "kimi-k3": {
    "scores": {
     "Consecutive events": 89.8,
     "Table join": 48.4,
     "Table reformat": 98.0
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "Consecutive events": 70.5,
     "Table join": 48.3,
     "Table reformat": 98.0
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "Consecutive events": 68.1,
     "Table join": 48.0,
     "Table reformat": 98.0
    }
   },
   "mistral-large-4": {
    "scores": {
     "Consecutive events": 80.5,
     "Table join": 49.1,
     "Table reformat": 100.0
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "Consecutive events": 90.5,
     "Table join": 48.2,
     "Table reformat": 100.0
    }
   },
   "muse-spark-open": {
    "scores": {
     "Consecutive events": 36.0,
     "Table join": 44.5,
     "Table reformat": 100.0
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "Consecutive events": 98.5,
     "Table join": 46.3,
     "Table reformat": 100.0
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "Consecutive events": 50.4,
     "Table join": 46.7,
     "Table reformat": 100.0
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "Consecutive events": 49.2,
     "Table join": 46.4,
     "Table reformat": 98.0
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Consecutive events": 84.9,
     "Table join": 48.8,
     "Table reformat": 96.1
    }
   },
   "qwen3.8-max": {
    "scores": {
     "Consecutive events": 87.1,
     "Table join": 50.1,
     "Table reformat": 98.0
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "Consecutive events": 50.3,
     "Table join": 45.1,
     "Table reformat": 100.0
    }
   },
   "step-5-preview": {
    "scores": {
     "Consecutive events": 42.4,
     "Table join": 50.1,
     "Table reformat": 98.0
    }
   }
  }
 },
 "Research": {
  "parts": [
   {
    "name": "Regulatory & Compliance",
    "gloss": "Rules, filings and compliance questions answered from primary sources.",
    "tasks": 7
   },
   {
    "name": "Legal",
    "gloss": "Statutes, cases and contractual questions with cited authority.",
    "tasks": 4
   },
   {
    "name": "Tax & Finance",
    "gloss": "Tax rules and financial figures, with the arithmetic shown.",
    "tasks": 5
   },
   {
    "name": "Strategy",
    "gloss": "Market and competitive analysis with sourced evidence.",
    "tasks": 3
   },
   {
    "name": "Science & Evidence",
    "gloss": "Scientific and empirical questions weighed on the published evidence.",
    "tasks": 3
   }
  ],
  "models": {
   "claude-fable-5.1": {
    "scores": {
     "Regulatory & Compliance": 86.3,
     "Legal": 93.5,
     "Tax & Finance": 75.4,
     "Strategy": 56.5,
     "Science & Evidence": 93.5
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Regulatory & Compliance": 80.4,
     "Legal": 85.1,
     "Tax & Finance": 73.1,
     "Strategy": 85.4,
     "Science & Evidence": 92.2
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Regulatory & Compliance": 87.2,
     "Legal": 99.9,
     "Tax & Finance": 82.0,
     "Strategy": 87.5,
     "Science & Evidence": 62.4
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Regulatory & Compliance": 89.2,
     "Legal": 95.1,
     "Tax & Finance": 60.5,
     "Strategy": 82.6,
     "Science & Evidence": 94.8
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Regulatory & Compliance": 86.2,
     "Legal": 83.7,
     "Tax & Finance": 60.0,
     "Strategy": 81.6,
     "Science & Evidence": 89.0
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "Regulatory & Compliance": 79.7,
     "Legal": 86.7,
     "Tax & Finance": 72.3,
     "Strategy": 80.4,
     "Science & Evidence": 77.6
    }
   },
   "gemini-4-argon": {
    "scores": {
     "Regulatory & Compliance": 77.0,
     "Legal": 98.1,
     "Tax & Finance": 76.5,
     "Strategy": 84.8,
     "Science & Evidence": 86.7
    }
   },
   "gemma-4-31b": {
    "scores": {
     "Regulatory & Compliance": 49.3,
     "Legal": 40.8,
     "Tax & Finance": 48.5,
     "Strategy": 47.7,
     "Science & Evidence": 27.2
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "Regulatory & Compliance": 72.1,
     "Legal": 69.3,
     "Tax & Finance": 56.0,
     "Strategy": 70.7,
     "Science & Evidence": 89.4
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "Regulatory & Compliance": 57.8,
     "Legal": 71.7,
     "Tax & Finance": 40.6,
     "Strategy": 56.0,
     "Science & Evidence": 88.6
    }
   },
   "gpt-6-astra": {
    "scores": {
     "Regulatory & Compliance": 80.0,
     "Legal": 94.4,
     "Tax & Finance": 85.0,
     "Strategy": 87.1,
     "Science & Evidence": 84.0
    }
   },
   "gpt-6-luna": {
    "scores": {
     "Regulatory & Compliance": 35.9,
     "Legal": 73.4,
     "Tax & Finance": 14.3,
     "Strategy": 55.3,
     "Science & Evidence": 58.3
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Regulatory & Compliance": 77.3,
     "Legal": 99.8,
     "Tax & Finance": 81.8,
     "Strategy": 76.0,
     "Science & Evidence": 86.7
    }
   },
   "gpt-6.1-terra": {
    "scores": {
     "Regulatory & Compliance": 62.8,
     "Legal": 85.6,
     "Tax & Finance": 67.8,
     "Strategy": 78.5,
     "Science & Evidence": 89.8
    }
   },
   "grok-4.6": {
    "scores": {
     "Regulatory & Compliance": 85.3,
     "Legal": 88.5,
     "Tax & Finance": 79.9,
     "Strategy": 82.9,
     "Science & Evidence": 86.8
    }
   },
   "grok-4.7": {
    "scores": {
     "Regulatory & Compliance": 83.9,
     "Legal": 90.8,
     "Tax & Finance": 81.9,
     "Strategy": 91.6,
     "Science & Evidence": 88.9
    }
   },
   "inkling": {
    "scores": {
     "Regulatory & Compliance": 57.8,
     "Legal": 60.7,
     "Tax & Finance": 60.0,
     "Strategy": 64.9,
     "Science & Evidence": 58.9
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Regulatory & Compliance": 79.9,
     "Legal": 99.1,
     "Tax & Finance": 62.4,
     "Strategy": 90.5,
     "Science & Evidence": 58.4
    }
   },
   "kimi-k3": {
    "scores": {
     "Regulatory & Compliance": 78.4,
     "Legal": 95.1,
     "Tax & Finance": 67.7,
     "Strategy": 69.2,
     "Science & Evidence": 87.5
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "Regulatory & Compliance": 77.0,
     "Legal": 77.4,
     "Tax & Finance": 64.3,
     "Strategy": 80.7,
     "Science & Evidence": 91.6
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "Regulatory & Compliance": 65.0,
     "Legal": 68.1,
     "Tax & Finance": 64.4,
     "Strategy": 88.8,
     "Science & Evidence": 94.3
    }
   },
   "mistral-large-4": {
    "scores": {
     "Regulatory & Compliance": 33.5,
     "Legal": 45.6,
     "Tax & Finance": 11.0,
     "Strategy": 48.7,
     "Science & Evidence": 0.0
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "Regulatory & Compliance": 87.9,
     "Legal": 98.4,
     "Tax & Finance": 28.5,
     "Strategy": 60.7,
     "Science & Evidence": 58.2
    }
   },
   "muse-spark-open": {
    "scores": {
     "Regulatory & Compliance": 76.7,
     "Legal": 90.6,
     "Tax & Finance": 72.6,
     "Strategy": 82.6,
     "Science & Evidence": 85.7
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "Regulatory & Compliance": 85.0,
     "Legal": 99.8,
     "Tax & Finance": 74.8,
     "Strategy": 97.0,
     "Science & Evidence": 92.1
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "Regulatory & Compliance": 60.1,
     "Legal": 89.2,
     "Tax & Finance": 65.8,
     "Strategy": 69.1,
     "Science & Evidence": 81.7
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "Regulatory & Compliance": 69.1,
     "Legal": 70.5,
     "Tax & Finance": 62.2,
     "Strategy": 73.8,
     "Science & Evidence": 56.7
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Regulatory & Compliance": 63.5,
     "Legal": 76.5,
     "Tax & Finance": 37.9,
     "Strategy": 59.4,
     "Science & Evidence": 62.8
    }
   },
   "qwen3.8-max": {
    "scores": {
     "Regulatory & Compliance": 73.0,
     "Legal": 90.1,
     "Tax & Finance": 57.8,
     "Strategy": 71.0,
     "Science & Evidence": 55.0
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "Regulatory & Compliance": 50.4,
     "Legal": 75.6,
     "Tax & Finance": 70.1,
     "Strategy": 81.7,
     "Science & Evidence": 79.9
    }
   },
   "step-5-preview": {
    "scores": {
     "Regulatory & Compliance": 67.3,
     "Legal": 69.6,
     "Tax & Finance": 71.2,
     "Strategy": 81.4,
     "Science & Evidence": 84.3
    }
   }
  }
 },
 "Design": {
  "parts": [
   {
    "name": "Documents (DOCX)",
    "gloss": "Word documents: layout, structure and formatting to a brief.",
    "tasks": 5
   },
   {
    "name": "PDF",
    "gloss": "PDF deliverables: typeset pages, forms and reports.",
    "tasks": 8
   },
   {
    "name": "Presentations (PPT)",
    "gloss": "Slide decks: visual design, content organization and fit to the request.",
    "tasks": 12
   }
  ],
  "models": {
   "claude-fable-5.1": {
    "scores": {
     "Documents (DOCX)": 92.1,
     "PDF": 84.2,
     "Presentations (PPT)": 79.2
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Documents (DOCX)": 92.4,
     "PDF": 87.3,
     "Presentations (PPT)": 76.4
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Documents (DOCX)": 90.3,
     "PDF": 88.2,
     "Presentations (PPT)": 84.6
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Documents (DOCX)": 94.1,
     "PDF": 87.6,
     "Presentations (PPT)": 79.4
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Documents (DOCX)": 81.3,
     "PDF": 89.2,
     "Presentations (PPT)": 80.4
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "Documents (DOCX)": 83.1,
     "PDF": 77.8,
     "Presentations (PPT)": 64.9
    }
   },
   "gemini-4-argon": {
    "scores": {
     "Documents (DOCX)": 84.5,
     "PDF": 74.6,
     "Presentations (PPT)": 68.6
    }
   },
   "gemma-4-31b": {
    "scores": {
     "Documents (DOCX)": 64.6,
     "PDF": 54.9,
     "Presentations (PPT)": 58.1
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "Documents (DOCX)": 72.7,
     "PDF": 78.9,
     "Presentations (PPT)": 78.0
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "Documents (DOCX)": 79.2,
     "PDF": 75.5,
     "Presentations (PPT)": 75.0
    }
   },
   "gpt-6-astra": {
    "scores": {
     "Documents (DOCX)": 84.7,
     "PDF": 78.3,
     "Presentations (PPT)": 83.4
    }
   },
   "gpt-6-luna": {
    "scores": {
     "Documents (DOCX)": 86.5,
     "PDF": 78.9,
     "Presentations (PPT)": 83.2
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Documents (DOCX)": 86.2,
     "PDF": 79.1,
     "Presentations (PPT)": 82.3
    }
   },
   "gpt-6.1-terra": {
    "scores": {
     "Documents (DOCX)": 93.0,
     "PDF": 75.5,
     "Presentations (PPT)": 82.5
    }
   },
   "grok-4.6": {
    "scores": {
     "Documents (DOCX)": 76.9,
     "PDF": 82.0,
     "Presentations (PPT)": 79.6
    }
   },
   "grok-4.7": {
    "scores": {
     "Documents (DOCX)": 72.9,
     "PDF": 80.1,
     "Presentations (PPT)": 78.6
    }
   },
   "inkling": {
    "scores": {
     "Documents (DOCX)": 72.5,
     "PDF": 50.5,
     "Presentations (PPT)": 57.5
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Documents (DOCX)": 78.8,
     "PDF": 82.5,
     "Presentations (PPT)": 74.2
    }
   },
   "kimi-k3": {
    "scores": {
     "Documents (DOCX)": 81.4,
     "PDF": 79.0,
     "Presentations (PPT)": 74.2
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "Documents (DOCX)": 86.5,
     "PDF": 80.0,
     "Presentations (PPT)": 74.1
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "Documents (DOCX)": 84.6,
     "PDF": 77.7,
     "Presentations (PPT)": 67.2
    }
   },
   "mistral-large-4": {
    "scores": {
     "Documents (DOCX)": 81.1,
     "PDF": 52.1,
     "Presentations (PPT)": 68.1
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "Documents (DOCX)": 73.4,
     "PDF": 81.1,
     "Presentations (PPT)": 71.1
    }
   },
   "muse-spark-open": {
    "scores": {
     "Documents (DOCX)": 61.3,
     "PDF": 67.1,
     "Presentations (PPT)": 55.3
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "Documents (DOCX)": 74.7,
     "PDF": 70.6,
     "Presentations (PPT)": 78.3
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "Documents (DOCX)": 64.4,
     "PDF": 59.3,
     "Presentations (PPT)": 52.6
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "Documents (DOCX)": 74.0,
     "PDF": 59.5,
     "Presentations (PPT)": 60.4
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Documents (DOCX)": 79.9,
     "PDF": 82.3,
     "Presentations (PPT)": 69.9
    }
   },
   "qwen3.8-max": {
    "scores": {
     "Documents (DOCX)": 78.7,
     "PDF": 80.0,
     "Presentations (PPT)": 77.5
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "Documents (DOCX)": 70.7,
     "PDF": 61.8,
     "Presentations (PPT)": 60.0
    }
   },
   "step-5-preview": {
    "scores": {
     "Documents (DOCX)": 82.7,
     "PDF": 72.9,
     "Presentations (PPT)": 77.1
    }
   }
  }
 }
}