{
 "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": 4
   },
   {
    "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": "Systems Engineering",
    "gloss": "Build and repair complex systems from a specification, then keep them running under realistic load and failures.",
    "tasks": 3
   },
   {
    "name": "Performance Optimisation",
    "gloss": "Make real programs faster and real schedules tighter, scored on measured results rather than judgement.",
    "tasks": 4
   }
  ],
  "models": {
   "beam-1.0": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 4.2,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 51.7,
     "Performance Optimisation": 35.0
    }
   },
   "claude-fable-5.1": {
    "scores": {
     "Finance & Compliance Apps": 95.8,
     "Operations & Commerce Apps": 81.2,
     "Long-Run Operations": 47.8,
     "Systems Engineering": 92.8,
     "Performance Optimisation": 81.0
    }
   },
   "claude-fable-5": {
    "scores": {
     "Finance & Compliance Apps": 58.4,
     "Operations & Commerce Apps": 47.9,
     "Long-Run Operations": 1.4,
     "Systems Engineering": 88.3,
     "Performance Optimisation": 65.9
    }
   },
   "claude-haiku-4.5": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 13.8,
     "Performance Optimisation": 39.6
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Finance & Compliance Apps": 70.8,
     "Operations & Commerce Apps": 43.8,
     "Long-Run Operations": 33.3,
     "Systems Engineering": 62.8,
     "Performance Optimisation": 72.7
    }
   },
   "claude-opus-4.8": {
    "scores": {
     "Finance & Compliance Apps": 54.1,
     "Operations & Commerce Apps": 31.2,
     "Long-Run Operations": 11.1,
     "Systems Engineering": 73.9,
     "Performance Optimisation": 68.3
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Finance & Compliance Apps": 100.0,
     "Operations & Commerce Apps": 87.5,
     "Long-Run Operations": 45.4,
     "Systems Engineering": 98.1,
     "Performance Optimisation": 87.6
    }
   },
   "claude-opus-5": {
    "scores": {
     "Finance & Compliance Apps": 62.5,
     "Operations & Commerce Apps": 60.4,
     "Long-Run Operations": 23.6,
     "Systems Engineering": 95.2,
     "Performance Optimisation": 67.3
    }
   },
   "claude-sonnet-4.6": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 10.4,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 68.2,
     "Performance Optimisation": 42.6
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Finance & Compliance Apps": 100.0,
     "Operations & Commerce Apps": 81.2,
     "Long-Run Operations": 36.1,
     "Systems Engineering": 84.9,
     "Performance Optimisation": 82.5
    }
   },
   "claude-sonnet-5": {
    "scores": {
     "Finance & Compliance Apps": 16.6,
     "Operations & Commerce Apps": 39.6,
     "Long-Run Operations": 5.6,
     "Systems Engineering": 70.1,
     "Performance Optimisation": 62.4
    }
   },
   "deepseek-v4-flash-vision": {
    "scores": {
     "Finance & Compliance Apps": 37.5,
     "Operations & Commerce Apps": 33.3,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 83.5,
     "Performance Optimisation": 42.0
    }
   },
   "deepseek-v4-flash": {
    "scores": {
     "Finance & Compliance Apps": 8.3,
     "Operations & Commerce Apps": 27.1,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 82.4,
     "Performance Optimisation": 67.5
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Finance & Compliance Apps": 41.7,
     "Operations & Commerce Apps": 37.5,
     "Long-Run Operations": 34.7,
     "Systems Engineering": 82.2,
     "Performance Optimisation": 62.0
    }
   },
   "deepseek-v4.1-flash": {
    "scores": {
     "Finance & Compliance Apps": 20.8,
     "Operations & Commerce Apps": 41.7,
     "Long-Run Operations": 1.4,
     "Systems Engineering": 87.2,
     "Performance Optimisation": 51.6
    }
   },
   "gemini-3.1-pro": {
    "scores": {
     "Finance & Compliance Apps": 8.3,
     "Operations & Commerce Apps": 4.2,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 43.3,
     "Performance Optimisation": 62.5
    }
   },
   "gemini-3.7-flash": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 8.3,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 71.2,
     "Performance Optimisation": 65.8
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "Finance & Compliance Apps": 16.7,
     "Operations & Commerce Apps": 8.3,
     "Long-Run Operations": 11.1,
     "Systems Engineering": 78.2,
     "Performance Optimisation": 43.5
    }
   },
   "gemini-4-argon": {
    "scores": {
     "Finance & Compliance Apps": 91.7,
     "Operations & Commerce Apps": 66.7,
     "Long-Run Operations": 19.4,
     "Systems Engineering": 97.6,
     "Performance Optimisation": 68.4
    }
   },
   "gemma-4-31b": {
    "scores": {
     "Finance & Compliance Apps": 4.2,
     "Operations & Commerce Apps": 8.3,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 19.6,
     "Performance Optimisation": 32.7
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "Finance & Compliance Apps": 8.3,
     "Operations & Commerce Apps": 20.9,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 79.0,
     "Performance Optimisation": 39.5
    }
   },
   "glm-5.3": {
    "scores": {
     "Finance & Compliance Apps": 37.5,
     "Operations & Commerce Apps": 27.1,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 65.0,
     "Performance Optimisation": 48.3
    }
   },
   "gpt-5.5": {
    "scores": {
     "Finance & Compliance Apps": 37.5,
     "Operations & Commerce Apps": 16.7,
     "Long-Run Operations": 22.2,
     "Systems Engineering": 81.4,
     "Performance Optimisation": 68.8
    }
   },
   "gpt-5.6-luna": {
    "scores": {
     "Finance & Compliance Apps": 12.5,
     "Operations & Commerce Apps": 12.5,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 69.5,
     "Performance Optimisation": 59.6
    }
   },
   "gpt-5.6-sol": {
    "scores": {
     "Finance & Compliance Apps": 54.2,
     "Operations & Commerce Apps": 41.7,
     "Long-Run Operations": 16.7,
     "Systems Engineering": 87.8,
     "Performance Optimisation": 44.2
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "Finance & Compliance Apps": 54.1,
     "Operations & Commerce Apps": 27.1,
     "Long-Run Operations": 11.1,
     "Systems Engineering": 79.1,
     "Performance Optimisation": 68.7
    }
   },
   "gpt-6-astra": {
    "scores": {
     "Finance & Compliance Apps": 100.0,
     "Operations & Commerce Apps": 100.0,
     "Long-Run Operations": 29.1,
     "Systems Engineering": 98.3,
     "Performance Optimisation": 75.1
    }
   },
   "gpt-6-luna": {
    "scores": {
     "Finance & Compliance Apps": 45.9,
     "Operations & Commerce Apps": 43.8,
     "Long-Run Operations": 15.3,
     "Systems Engineering": 67.8,
     "Performance Optimisation": 64.4
    }
   },
   "gpt-6-sol": {
    "scores": {
     "Finance & Compliance Apps": 83.3,
     "Operations & Commerce Apps": 54.2,
     "Long-Run Operations": 38.9,
     "Systems Engineering": 85.2,
     "Performance Optimisation": 70.5
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Finance & Compliance Apps": 66.7,
     "Operations & Commerce Apps": 93.8,
     "Long-Run Operations": 20.8,
     "Systems Engineering": 94.6,
     "Performance Optimisation": 81.5
    }
   },
   "grok-4.5": {
    "scores": {
     "Finance & Compliance Apps": 12.5,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 67.6,
     "Performance Optimisation": 62.4
    }
   },
   "grok-4.6": {
    "scores": {
     "Finance & Compliance Apps": 37.5,
     "Operations & Commerce Apps": 35.4,
     "Long-Run Operations": 5.6,
     "Systems Engineering": 80.4,
     "Performance Optimisation": 45.6
    }
   },
   "grok-4.7": {
    "scores": {
     "Finance & Compliance Apps": 50.0,
     "Operations & Commerce Apps": 50.0,
     "Long-Run Operations": 5.6,
     "Systems Engineering": 79.3,
     "Performance Optimisation": 61.7
    }
   },
   "inkling": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 38.0,
     "Performance Optimisation": 34.9
    }
   },
   "kimi-k3": {
    "scores": {
     "Finance & Compliance Apps": 37.5,
     "Operations & Commerce Apps": 8.3,
     "Long-Run Operations": 6.9,
     "Systems Engineering": 86.4,
     "Performance Optimisation": 76.8
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "Finance & Compliance Apps": 29.2,
     "Operations & Commerce Apps": 43.8,
     "Long-Run Operations": 13.9,
     "Systems Engineering": 81.5,
     "Performance Optimisation": 69.3
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "Finance & Compliance Apps": 12.5,
     "Operations & Commerce Apps": 12.5,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 77.1,
     "Performance Optimisation": 73.5
    }
   },
   "minimax-m3": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 29.4,
     "Performance Optimisation": 34.0
    }
   },
   "mistral-large-4": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 19.6,
     "Performance Optimisation": 38.2
    }
   },
   "muse-spark-1.2": {
    "scores": {
     "Finance & Compliance Apps": 4.2,
     "Operations & Commerce Apps": 4.2,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 67.9,
     "Performance Optimisation": 36.7
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "Finance & Compliance Apps": 29.2,
     "Operations & Commerce Apps": 33.3,
     "Long-Run Operations": 6.9,
     "Systems Engineering": 69.9,
     "Performance Optimisation": 61.5
    }
   },
   "muse-spark-open": {
    "scores": {
     "Finance & Compliance Apps": 12.5,
     "Operations & Commerce Apps": 4.2,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 44.2,
     "Performance Optimisation": 35.3
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "Finance & Compliance Apps": 45.8,
     "Operations & Commerce Apps": 45.8,
     "Long-Run Operations": 20.8,
     "Systems Engineering": 88.8,
     "Performance Optimisation": 65.4
    }
   },
   "nemotron-lightning-3.5": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 13.4,
     "Performance Optimisation": 27.5
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 27.0,
     "Performance Optimisation": 31.6
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "Finance & Compliance Apps": 4.2,
     "Operations & Commerce Apps": 4.2,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 35.7,
     "Performance Optimisation": 35.8
    }
   },
   "nemotron-ultra-3": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 22.7,
     "Performance Optimisation": 33.3
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Finance & Compliance Apps": 8.3,
     "Operations & Commerce Apps": 25.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 67.3,
     "Performance Optimisation": 41.6
    }
   },
   "qwen3.8-flash": {
    "scores": {
     "Finance & Compliance Apps": 12.5,
     "Operations & Commerce Apps": 45.8,
     "Long-Run Operations": 18.1,
     "Systems Engineering": 65.3,
     "Performance Optimisation": 46.2
    }
   },
   "qwen3.8-max": {
    "scores": {
     "Finance & Compliance Apps": 29.1,
     "Operations & Commerce Apps": 22.9,
     "Long-Run Operations": 12.5,
     "Systems Engineering": 81.4,
     "Performance Optimisation": 67.8
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "Finance & Compliance Apps": 0.0,
     "Operations & Commerce Apps": 0.0,
     "Long-Run Operations": 0.0,
     "Systems Engineering": 37.2,
     "Performance Optimisation": 60.1
    }
   },
   "step-5-preview": {
    "scores": {
     "Finance & Compliance Apps": 4.2,
     "Operations & Commerce Apps": 16.7,
     "Long-Run Operations": 5.6,
     "Systems Engineering": 75.2,
     "Performance Optimisation": 42.5
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Finance & Compliance Apps": 62.5,
     "Operations & Commerce Apps": 20.8,
     "Long-Run Operations": 25.0,
     "Systems Engineering": 96.4,
     "Performance Optimisation": 48.4
    }
   }
  }
 },
 "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": {
   "beam-1.0": {
    "scores": {
     "Theory of mind": 76.9,
     "Zebra puzzles": 82.0,
     "Spatial": 94.0,
     "Logic with navigation": 68.0
    }
   },
   "claude-fable-5.1": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 86.0
    }
   },
   "claude-fable-5": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 100.0,
     "Spatial": 96.0,
     "Logic with navigation": 78.0
    }
   },
   "claude-haiku-4.5": {
    "scores": {
     "Theory of mind": 44.2,
     "Zebra puzzles": 58.5,
     "Spatial": 90.0,
     "Logic with navigation": 54.0
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Theory of mind": 78.8,
     "Zebra puzzles": 96.8,
     "Spatial": 88.0,
     "Logic with navigation": 66.0
    }
   },
   "claude-opus-4.8": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 78.0
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 86.0
    }
   },
   "claude-opus-5": {
    "scores": {
     "Theory of mind": 78.8,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 86.0
    }
   },
   "claude-sonnet-4.6": {
    "scores": {
     "Theory of mind": 75.0,
     "Zebra puzzles": 94.5,
     "Spatial": 100.0,
     "Logic with navigation": 76.0
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 82.0
    }
   },
   "claude-sonnet-5": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 88.0,
     "Spatial": 100.0,
     "Logic with navigation": 86.0
    }
   },
   "deepseek-v4-flash-vision": {
    "scores": {
     "Theory of mind": 78.8,
     "Zebra puzzles": 98.8,
     "Spatial": 96.0,
     "Logic with navigation": 68.0
    }
   },
   "deepseek-v4-flash": {
    "scores": {
     "Theory of mind": 73.1,
     "Zebra puzzles": 49.2,
     "Spatial": 96.0,
     "Logic with navigation": 64.0
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Theory of mind": 88.5,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 70.0
    }
   },
   "deepseek-v4.1-flash": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 68.0
    }
   },
   "gemini-3.1-pro": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 85.2,
     "Spatial": 98.0,
     "Logic with navigation": 72.0
    }
   },
   "gemini-3.7-flash": {
    "scores": {
     "Theory of mind": 82.7,
     "Zebra puzzles": 94.5,
     "Spatial": 100.0,
     "Logic with navigation": 74.0
    }
   },
   "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
    }
   },
   "glm-5.3": {
    "scores": {
     "Theory of mind": 88.5,
     "Zebra puzzles": 88.8,
     "Spatial": 96.0,
     "Logic with navigation": 70.0
    }
   },
   "gpt-5.5": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 76.0
    }
   },
   "gpt-5.6-luna": {
    "scores": {
     "Theory of mind": 73.1,
     "Zebra puzzles": 93.5,
     "Spatial": 96.0,
     "Logic with navigation": 80.0
    }
   },
   "gpt-5.6-sol": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 82.0
    }
   },
   "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-sol": {
    "scores": {
     "Theory of mind": 84.6,
     "Zebra puzzles": 100.0,
     "Spatial": 96.0,
     "Logic with navigation": 74.0
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 100.0,
     "Spatial": 98.0,
     "Logic with navigation": 86.0
    }
   },
   "grok-4.5": {
    "scores": {
     "Theory of mind": 82.7,
     "Zebra puzzles": 94.0,
     "Spatial": 100.0,
     "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": {
    "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
    }
   },
   "minimax-m3": {
    "scores": {
     "Theory of mind": 76.9,
     "Zebra puzzles": 61.0,
     "Spatial": 94.0,
     "Logic with navigation": 66.0
    }
   },
   "mistral-large-4": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 93.0,
     "Spatial": 94.0,
     "Logic with navigation": 68.0
    }
   },
   "muse-spark-1.2": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 95.2,
     "Spatial": 96.0,
     "Logic with navigation": 88.0
    }
   },
   "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-lightning-3.5": {
    "scores": {
     "Theory of mind": 44.2,
     "Zebra puzzles": 17.2,
     "Spatial": 92.0,
     "Logic with navigation": 18.0
    }
   },
   "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": 76.9,
     "Zebra puzzles": 80.2,
     "Spatial": 98.0,
     "Logic with navigation": 60.0
    }
   },
   "nemotron-ultra-3": {
    "scores": {
     "Theory of mind": 61.5,
     "Zebra puzzles": 71.2,
     "Spatial": 98.0,
     "Logic with navigation": 68.0
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Theory of mind": 59.6,
     "Zebra puzzles": 96.5,
     "Spatial": 96.0,
     "Logic with navigation": 68.0
    }
   },
   "qwen3.8-flash": {
    "scores": {
     "Theory of mind": 80.8,
     "Zebra puzzles": 98.8,
     "Spatial": 100.0,
     "Logic with navigation": 70.0
    }
   },
   "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
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Theory of mind": 86.5,
     "Zebra puzzles": 100.0,
     "Spatial": 100.0,
     "Logic with navigation": 78.0
    }
   }
  }
 },
 "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": {
   "beam-1.0": {
    "scores": {
     "LLM API Integration": 87.5,
     "Agent Loop": 80.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 91.7,
     "Product & UX": 55.6,
     "Memory & Tools": 100.0
    }
   },
   "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-fable-5": {
    "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-4.5": {
    "scores": {
     "LLM API Integration": 62.5,
     "Agent Loop": 50.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 50.0,
     "Product & UX": 16.7,
     "Memory & Tools": 74.0
    }
   },
   "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-4.8": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 60.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 66.7,
     "Memory & Tools": 50.0
    }
   },
   "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-opus-5": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 66.7,
     "Memory & Tools": 75.0
    }
   },
   "claude-sonnet-4.6": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 40.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 66.7,
     "Product & UX": 50.0,
     "Memory & Tools": 75.0
    }
   },
   "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
    }
   },
   "claude-sonnet-5": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 80.0,
     "Billing & Limits": 79.2,
     "Cloud Infrastructure": 91.7,
     "Product & UX": 66.7,
     "Memory & Tools": 75.0
    }
   },
   "deepseek-v4-flash-vision": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 40.0,
     "Billing & Limits": 25.0,
     "Cloud Infrastructure": 48.3,
     "Product & UX": 0.0,
     "Memory & Tools": 50.0
    }
   },
   "deepseek-v4-flash": {
    "scores": {
     "LLM API Integration": 58.3,
     "Agent Loop": 80.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 86.1,
     "Product & UX": 72.2,
     "Memory & Tools": 87.5
    }
   },
   "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
    }
   },
   "deepseek-v4.1-flash": {
    "scores": {
     "LLM API Integration": 87.5,
     "Agent Loop": 100.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 88.9,
     "Product & UX": 88.9,
     "Memory & Tools": 100.0
    }
   },
   "gemini-3.1-pro": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 80.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 66.7,
     "Product & UX": 50.0,
     "Memory & Tools": 75.0
    }
   },
   "gemini-3.7-flash": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 80.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 83.3,
     "Product & UX": 66.7,
     "Memory & Tools": 75.0
    }
   },
   "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
    }
   },
   "glm-5.3": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 83.3,
     "Product & UX": 66.7,
     "Memory & Tools": 50.0
    }
   },
   "gpt-5.5": {
    "scores": {
     "LLM API Integration": 87.5,
     "Agent Loop": 100.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 75.0,
     "Product & UX": 77.8,
     "Memory & Tools": 100.0
    }
   },
   "gpt-5.6-luna": {
    "scores": {
     "LLM API Integration": 87.5,
     "Agent Loop": 80.0,
     "Billing & Limits": 29.2,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 61.1,
     "Memory & Tools": 75.0
    }
   },
   "gpt-5.6-sol": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 66.7,
     "Memory & Tools": 100.0
    }
   },
   "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-sol": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 100.0,
     "Product & UX": 66.7,
     "Memory & Tools": 100.0
    }
   },
   "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
    }
   },
   "grok-4.5": {
    "scores": {
     "LLM API Integration": 75.0,
     "Agent Loop": 100.0,
     "Billing & Limits": 100.0,
     "Cloud Infrastructure": 94.4,
     "Product & UX": 94.4,
     "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": {
    "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
    }
   },
   "minimax-m3": {
    "scores": {
     "LLM API Integration": 50.0,
     "Agent Loop": 50.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 88.9,
     "Product & UX": 83.3,
     "Memory & Tools": 92.9
    }
   },
   "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-1.2": {
    "scores": {
     "LLM API Integration": 62.5,
     "Agent Loop": 60.0,
     "Billing & Limits": 75.0,
     "Cloud Infrastructure": 91.7,
     "Product & UX": 77.8,
     "Memory & Tools": 100.0
    }
   },
   "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-lightning-3.5": {
    "scores": {
     "LLM API Integration": 12.5,
     "Agent Loop": 20.0,
     "Billing & Limits": 25.0,
     "Cloud Infrastructure": 8.3,
     "Product & UX": 22.2,
     "Memory & Tools": 55.4
    }
   },
   "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": 75.0,
     "Agent Loop": 70.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 75.0,
     "Product & UX": 33.3,
     "Memory & Tools": 100.0
    }
   },
   "nemotron-ultra-3": {
    "scores": {
     "LLM API Integration": 45.8,
     "Agent Loop": 40.0,
     "Billing & Limits": 50.0,
     "Cloud Infrastructure": 55.6,
     "Product & UX": 0.0,
     "Memory & Tools": 86.6
    }
   },
   "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-flash": {
    "scores": {
     "LLM API Integration": 100.0,
     "Agent Loop": 80.0,
     "Billing & Limits": 95.8,
     "Cloud Infrastructure": 91.7,
     "Product & UX": 66.7,
     "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
    }
   },
   "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
    }
   }
  }
 },
 "Browser": {
  "parts": [
   {
    "name": "Data Aggregation",
    "gloss": "Exhaustive traversal of a large record set and multi-part roll-ups.",
    "tasks": 7
   },
   {
    "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": 6
   },
   {
    "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": {
   "beam-1.0": {
    "scores": {
     "Data Aggregation": 0.0,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "claude-fable-5.1": {
    "scores": {
     "Data Aggregation": 100.0,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "claude-fable-5": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "claude-haiku-4.5": {
    "scores": {
     "Data Aggregation": 14.3,
     "Sequential Planning": 50.0,
     "Analysis-Driven Action": 50.0,
     "Structural Integrity": 40.0,
     "Constraint Adherence": 66.7
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 75.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 66.7
    }
   },
   "claude-opus-4.8": {
    "scores": {
     "Data Aggregation": 85.7,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 50.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "claude-opus-5": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "claude-sonnet-4.6": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 40.0,
     "Constraint Adherence": 66.7
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 50.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "claude-sonnet-5": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 50.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 66.7
    }
   },
   "deepseek-v4-flash-vision": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 75.0,
     "Analysis-Driven Action": 50.0,
     "Structural Integrity": 60.0,
     "Constraint Adherence": 66.7
    }
   },
   "deepseek-v4-flash": {
    "scores": {
     "Data Aggregation": 0.0,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 66.7,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 66.7
    }
   },
   "deepseek-v4.1-flash": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 60.0,
     "Constraint Adherence": 100.0
    }
   },
   "gemini-3.1-pro": {
    "scores": {
     "Data Aggregation": 14.3,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "gemini-3.7-flash": {
    "scores": {
     "Data Aggregation": 42.9,
     "Sequential Planning": 50.0,
     "Analysis-Driven Action": 33.3,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "gemini-3.8-flash": {
    "scores": {
     "Data Aggregation": 42.9,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "gemini-4-argon": {
    "scores": {
     "Data Aggregation": 85.7,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "gemma-4-31b": {
    "scores": {
     "Data Aggregation": 14.3,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "glm-5.3-flash": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "glm-5.3": {
    "scores": {
     "Data Aggregation": 0.0,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "gpt-5.5": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 20.0,
     "Constraint Adherence": 100.0
    }
   },
   "gpt-5.6-luna": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 75.0,
     "Analysis-Driven Action": 66.7,
     "Structural Integrity": 40.0,
     "Constraint Adherence": 100.0
    }
   },
   "gpt-5.6-sol": {
    "scores": {
     "Data Aggregation": 86.0,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 83.0,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "gpt-5.6-terra": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "gpt-6-astra": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 66.7
    }
   },
   "gpt-6-luna": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 75.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 40.0,
     "Constraint Adherence": 100.0
    }
   },
   "gpt-6-sol": {
    "scores": {
     "Data Aggregation": 85.7,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Data Aggregation": 100.0,
     "Sequential Planning": 100.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 66.7
    }
   },
   "grok-4.5": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 60.0,
     "Constraint Adherence": 100.0
    }
   },
   "grok-4.6": {
    "scores": {
     "Data Aggregation": 42.9,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "grok-4.7": {
    "scores": {
     "Data Aggregation": 85.7,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 16.7,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 66.7
    }
   },
   "inkling": {
    "scores": {
     "Data Aggregation": 42.9,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 16.7,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "kimi-k3": {
    "scores": {
     "Data Aggregation": 28.6,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 20.0,
     "Constraint Adherence": 0.0
    }
   },
   "mimo-v2.6-pro": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 75.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "minimax-m3.1-flash": {
    "scores": {
     "Data Aggregation": 85.7,
     "Sequential Planning": 75.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 100.0,
     "Constraint Adherence": 100.0
    }
   },
   "minimax-m3": {
    "scores": {
     "Data Aggregation": 28.6,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "mistral-large-4": {
    "scores": {
     "Data Aggregation": 28.6,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "muse-spark-1.2": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 16.7,
     "Structural Integrity": 60.0,
     "Constraint Adherence": 33.3
    }
   },
   "muse-spark-1.3": {
    "scores": {
     "Data Aggregation": 71.4,
     "Sequential Planning": 50.0,
     "Analysis-Driven Action": 100.0,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "muse-spark-open": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 16.7,
     "Structural Integrity": 20.0,
     "Constraint Adherence": 66.7
    }
   },
   "muse-wonder-1.0": {
    "scores": {
     "Data Aggregation": 42.9,
     "Sequential Planning": 75.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 100.0
    }
   },
   "nemotron-lightning-3.5": {
    "scores": {
     "Data Aggregation": 0.0,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "nemotron-super-vl-3.5-nvfp4": {
    "scores": {
     "Data Aggregation": 28.6,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 20.0,
     "Constraint Adherence": 0.0
    }
   },
   "nemotron-super-vl-3.5": {
    "scores": {
     "Data Aggregation": 14.3,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "nemotron-ultra-3": {
    "scores": {
     "Data Aggregation": 0.0,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Data Aggregation": 57.1,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 80.0,
     "Constraint Adherence": 0.0
    }
   },
   "qwen3.8-flash": {
    "scores": {
     "Data Aggregation": 42.9,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 50.0,
     "Structural Integrity": 60.0,
     "Constraint Adherence": 66.7
    }
   },
   "qwen3.8-max": {
    "scores": {
     "Data Aggregation": 42.9,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 50.0,
     "Structural Integrity": 20.0,
     "Constraint Adherence": 33.3
    }
   },
   "smaug-nemotron-3.5-nvfp4": {
    "scores": {
     "Data Aggregation": 28.6,
     "Sequential Planning": 0.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 0.0,
     "Constraint Adherence": 0.0
    }
   },
   "step-5-preview": {
    "scores": {
     "Data Aggregation": 28.6,
     "Sequential Planning": 25.0,
     "Analysis-Driven Action": 0.0,
     "Structural Integrity": 40.0,
     "Constraint Adherence": 33.3
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Data Aggregation": 85.7,
     "Sequential Planning": 75.0,
     "Analysis-Driven Action": 83.3,
     "Structural Integrity": 60.0,
     "Constraint Adherence": 100.0
    }
   }
  }
 },
 "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": {
   "beam-1.0": {
    "scores": {
     "Consecutive events": 52.4,
     "Table join": 46.8,
     "Table reformat": 100.0
    }
   },
   "claude-fable-5.1": {
    "scores": {
     "Consecutive events": 90.6,
     "Table join": 56.2,
     "Table reformat": 94.1
    }
   },
   "claude-fable-5": {
    "scores": {
     "Consecutive events": 91.4,
     "Table join": 56.1,
     "Table reformat": 94.1
    }
   },
   "claude-haiku-4.5": {
    "scores": {
     "Consecutive events": 39.2,
     "Table join": 42.6,
     "Table reformat": 96.1
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Consecutive events": 16.4,
     "Table join": 44.8,
     "Table reformat": 98.0
    }
   },
   "claude-opus-4.8": {
    "scores": {
     "Consecutive events": 52.9,
     "Table join": 51.1,
     "Table reformat": 94.1
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Consecutive events": 90.6,
     "Table join": 52.3,
     "Table reformat": 98.0
    }
   },
   "claude-opus-5": {
    "scores": {
     "Consecutive events": 77.6,
     "Table join": 52.0,
     "Table reformat": 94.1
    }
   },
   "claude-sonnet-4.6": {
    "scores": {
     "Consecutive events": 82.9,
     "Table join": 49.2,
     "Table reformat": 96.1
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Consecutive events": 40.6,
     "Table join": 39.8,
     "Table reformat": 98.0
    }
   },
   "claude-sonnet-5": {
    "scores": {
     "Consecutive events": 74.7,
     "Table join": 42.4,
     "Table reformat": 98.0
    }
   },
   "deepseek-v4-flash-vision": {
    "scores": {
     "Consecutive events": 90.0,
     "Table join": 50.3,
     "Table reformat": 98.0
    }
   },
   "deepseek-v4-flash": {
    "scores": {
     "Consecutive events": 59.9,
     "Table join": 46.1,
     "Table reformat": 98.0
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Consecutive events": 90.3,
     "Table join": 48.2,
     "Table reformat": 100.0
    }
   },
   "deepseek-v4.1-flash": {
    "scores": {
     "Consecutive events": 88.7,
     "Table join": 49.0,
     "Table reformat": 100.0
    }
   },
   "gemini-3.1-pro": {
    "scores": {
     "Consecutive events": 85.2,
     "Table join": 52.3,
     "Table reformat": 98.0
    }
   },
   "gemini-3.7-flash": {
    "scores": {
     "Consecutive events": 60.4,
     "Table join": 45.5,
     "Table reformat": 98.0
    }
   },
   "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
    }
   },
   "glm-5.3": {
    "scores": {
     "Consecutive events": 72.3,
     "Table join": 42.3,
     "Table reformat": 96.1
    }
   },
   "gpt-5.5": {
    "scores": {
     "Consecutive events": 88.8,
     "Table join": 55.9,
     "Table reformat": 100.0
    }
   },
   "gpt-5.6-luna": {
    "scores": {
     "Consecutive events": 86.5,
     "Table join": 47.6,
     "Table reformat": 100.0
    }
   },
   "gpt-5.6-sol": {
    "scores": {
     "Consecutive events": 90.2,
     "Table join": 49.3,
     "Table reformat": 100.0
    }
   },
   "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-sol": {
    "scores": {
     "Consecutive events": 89.5,
     "Table join": 54.0,
     "Table reformat": 100.0
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Consecutive events": 90.6,
     "Table join": 57.4,
     "Table reformat": 100.0
    }
   },
   "grok-4.5": {
    "scores": {
     "Consecutive events": 75.5,
     "Table join": 43.6,
     "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": {
    "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
    }
   },
   "minimax-m3": {
    "scores": {
     "Consecutive events": 84.6,
     "Table join": 45.8,
     "Table reformat": 98.0
    }
   },
   "mistral-large-4": {
    "scores": {
     "Consecutive events": 80.5,
     "Table join": 49.1,
     "Table reformat": 100.0
    }
   },
   "muse-spark-1.2": {
    "scores": {
     "Consecutive events": 82.5,
     "Table join": 46.8,
     "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-lightning-3.5": {
    "scores": {
     "Consecutive events": 2.5,
     "Table join": 40.5,
     "Table reformat": 86.3
    }
   },
   "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": 49.6,
     "Table join": 44.3,
     "Table reformat": 100.0
    }
   },
   "nemotron-ultra-3": {
    "scores": {
     "Consecutive events": 18.8,
     "Table join": 46.6,
     "Table reformat": 98.0
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Consecutive events": 84.9,
     "Table join": 48.8,
     "Table reformat": 96.1
    }
   },
   "qwen3.8-flash": {
    "scores": {
     "Consecutive events": 75.6,
     "Table join": 51.0,
     "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
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Consecutive events": 90.2,
     "Table join": 49.1,
     "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": {
   "beam-1.0": {
    "scores": {
     "Regulatory & Compliance": 71.8,
     "Legal": 66.8,
     "Tax & Finance": 66.1,
     "Strategy": 78.8,
     "Science & Evidence": 88.6
    }
   },
   "claude-fable-5.1": {
    "scores": {
     "Regulatory & Compliance": 86.3,
     "Legal": 93.5,
     "Tax & Finance": 75.4,
     "Strategy": 56.5,
     "Science & Evidence": 93.5
    }
   },
   "claude-fable-5": {
    "scores": {
     "Regulatory & Compliance": 88.8,
     "Legal": 98.9,
     "Tax & Finance": 68.2,
     "Strategy": 90.7,
     "Science & Evidence": 93.6
    }
   },
   "claude-haiku-4.5": {
    "scores": {
     "Regulatory & Compliance": 43.5,
     "Legal": 48.0,
     "Tax & Finance": 45.3,
     "Strategy": 39.4,
     "Science & Evidence": 29.1
    }
   },
   "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-4.8": {
    "scores": {
     "Regulatory & Compliance": 86.7,
     "Legal": 92.5,
     "Tax & Finance": 74.4,
     "Strategy": 91.9,
     "Science & Evidence": 58.1
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Regulatory & Compliance": 87.2,
     "Legal": 99.9,
     "Tax & Finance": 82.0,
     "Strategy": 87.5,
     "Science & Evidence": 62.4
    }
   },
   "claude-opus-5": {
    "scores": {
     "Regulatory & Compliance": 89.2,
     "Legal": 97.2,
     "Tax & Finance": 79.4,
     "Strategy": 90.6,
     "Science & Evidence": 91.2
    }
   },
   "claude-sonnet-4.6": {
    "scores": {
     "Regulatory & Compliance": 71.8,
     "Legal": 77.1,
     "Tax & Finance": 60.2,
     "Strategy": 42.8,
     "Science & Evidence": 53.0
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Regulatory & Compliance": 89.2,
     "Legal": 95.1,
     "Tax & Finance": 60.5,
     "Strategy": 82.6,
     "Science & Evidence": 94.8
    }
   },
   "claude-sonnet-5": {
    "scores": {
     "Regulatory & Compliance": 76.4,
     "Legal": 97.1,
     "Tax & Finance": 74.2,
     "Strategy": 87.0,
     "Science & Evidence": 89.2
    }
   },
   "deepseek-v4-flash-vision": {
    "scores": {
     "Regulatory & Compliance": 80.6,
     "Legal": 83.0,
     "Tax & Finance": 36.2,
     "Strategy": 67.0,
     "Science & Evidence": 81.7
    }
   },
   "deepseek-v4-flash": {
    "scores": {
     "Regulatory & Compliance": 86.2,
     "Legal": 82.0,
     "Tax & Finance": 71.3,
     "Strategy": 70.7,
     "Science & Evidence": 80.9
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Regulatory & Compliance": 86.2,
     "Legal": 83.7,
     "Tax & Finance": 60.0,
     "Strategy": 81.6,
     "Science & Evidence": 89.0
    }
   },
   "deepseek-v4.1-flash": {
    "scores": {
     "Regulatory & Compliance": 84.8,
     "Legal": 91.0,
     "Tax & Finance": 84.6,
     "Strategy": 85.5,
     "Science & Evidence": 90.6
    }
   },
   "gemini-3.1-pro": {
    "scores": {
     "Regulatory & Compliance": 68.6,
     "Legal": 71.6,
     "Tax & Finance": 61.8,
     "Strategy": 68.8,
     "Science & Evidence": 54.6
    }
   },
   "gemini-3.7-flash": {
    "scores": {
     "Regulatory & Compliance": 83.3,
     "Legal": 89.7,
     "Tax & Finance": 63.6,
     "Strategy": 89.4,
     "Science & Evidence": 51.7
    }
   },
   "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
    }
   },
   "glm-5.3": {
    "scores": {
     "Regulatory & Compliance": 89.8,
     "Legal": 74.1,
     "Tax & Finance": 32.0,
     "Strategy": 93.2,
     "Science & Evidence": 81.6
    }
   },
   "gpt-5.5": {
    "scores": {
     "Regulatory & Compliance": 66.7,
     "Legal": 74.2,
     "Tax & Finance": 53.6,
     "Strategy": 55.1,
     "Science & Evidence": 52.5
    }
   },
   "gpt-5.6-luna": {
    "scores": {
     "Regulatory & Compliance": 65.6,
     "Legal": 25.0,
     "Tax & Finance": 29.3,
     "Strategy": 83.6,
     "Science & Evidence": 55.3
    }
   },
   "gpt-5.6-sol": {
    "scores": {
     "Regulatory & Compliance": 66.2,
     "Legal": 80.9,
     "Tax & Finance": 31.2,
     "Strategy": 58.9,
     "Science & Evidence": 26.2
    }
   },
   "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-sol": {
    "scores": {
     "Regulatory & Compliance": 83.8,
     "Legal": 89.3,
     "Tax & Finance": 46.7,
     "Strategy": 57.9,
     "Science & Evidence": 57.4
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Regulatory & Compliance": 77.3,
     "Legal": 99.8,
     "Tax & Finance": 81.8,
     "Strategy": 76.0,
     "Science & Evidence": 86.7
    }
   },
   "grok-4.5": {
    "scores": {
     "Regulatory & Compliance": 83.5,
     "Legal": 89.5,
     "Tax & Finance": 83.5,
     "Strategy": 90.5,
     "Science & Evidence": 86.3
    }
   },
   "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": {
    "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
    }
   },
   "minimax-m3": {
    "scores": {
     "Regulatory & Compliance": 73.7,
     "Legal": 63.6,
     "Tax & Finance": 62.7,
     "Strategy": 76.9,
     "Science & Evidence": 60.1
    }
   },
   "mistral-large-4": {
    "scores": {
     "Regulatory & Compliance": 33.5,
     "Legal": 45.6,
     "Tax & Finance": 11.0,
     "Strategy": 48.7,
     "Science & Evidence": 0.0
    }
   },
   "muse-spark-1.2": {
    "scores": {
     "Regulatory & Compliance": 73.3,
     "Legal": 96.6,
     "Tax & Finance": 50.1,
     "Strategy": 55.4,
     "Science & Evidence": 57.5
    }
   },
   "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-lightning-3.5": {
    "scores": {
     "Regulatory & Compliance": 42.4,
     "Legal": 29.9,
     "Tax & Finance": 38.7,
     "Strategy": 36.1,
     "Science & Evidence": 53.8
    }
   },
   "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": 64.4,
     "Legal": 77.5,
     "Tax & Finance": 69.5,
     "Strategy": 65.6,
     "Science & Evidence": 84.9
    }
   },
   "nemotron-ultra-3": {
    "scores": {
     "Regulatory & Compliance": 44.5,
     "Legal": 66.6,
     "Tax & Finance": 25.3,
     "Strategy": 55.5,
     "Science & Evidence": 26.2
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Regulatory & Compliance": 63.5,
     "Legal": 76.5,
     "Tax & Finance": 37.9,
     "Strategy": 59.4,
     "Science & Evidence": 62.8
    }
   },
   "qwen3.8-flash": {
    "scores": {
     "Regulatory & Compliance": 68.2,
     "Legal": 92.6,
     "Tax & Finance": 71.9,
     "Strategy": 54.5,
     "Science & Evidence": 90.3
    }
   },
   "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
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Regulatory & Compliance": 79.9,
     "Legal": 99.1,
     "Tax & Finance": 62.4,
     "Strategy": 90.5,
     "Science & Evidence": 58.4
    }
   }
  }
 },
 "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": {
   "beam-1.0": {
    "scores": {
     "Documents (DOCX)": 68.3,
     "PDF": 62.5,
     "Presentations (PPT)": 71.2
    }
   },
   "claude-fable-5.1": {
    "scores": {
     "Documents (DOCX)": 92.1,
     "PDF": 84.2,
     "Presentations (PPT)": 79.2
    }
   },
   "claude-fable-5": {
    "scores": {
     "Documents (DOCX)": 86.2,
     "PDF": 86.2,
     "Presentations (PPT)": 83.1
    }
   },
   "claude-haiku-4.5": {
    "scores": {
     "Documents (DOCX)": 70.2,
     "PDF": 54.6,
     "Presentations (PPT)": 59.1
    }
   },
   "claude-haiku-5.5": {
    "scores": {
     "Documents (DOCX)": 92.4,
     "PDF": 87.3,
     "Presentations (PPT)": 76.4
    }
   },
   "claude-opus-4.8": {
    "scores": {
     "Documents (DOCX)": 81.0,
     "PDF": 78.6,
     "Presentations (PPT)": 78.1
    }
   },
   "claude-opus-5.5": {
    "scores": {
     "Documents (DOCX)": 90.3,
     "PDF": 88.2,
     "Presentations (PPT)": 84.6
    }
   },
   "claude-opus-5": {
    "scores": {
     "Documents (DOCX)": 84.1,
     "PDF": 86.4,
     "Presentations (PPT)": 80.3
    }
   },
   "claude-sonnet-4.6": {
    "scores": {
     "Documents (DOCX)": 66.8,
     "PDF": 70.0,
     "Presentations (PPT)": 75.1
    }
   },
   "claude-sonnet-5.5": {
    "scores": {
     "Documents (DOCX)": 94.1,
     "PDF": 87.6,
     "Presentations (PPT)": 79.4
    }
   },
   "claude-sonnet-5": {
    "scores": {
     "Documents (DOCX)": 79.2,
     "PDF": 81.5,
     "Presentations (PPT)": 83.5
    }
   },
   "deepseek-v4-flash-vision": {
    "scores": {
     "Documents (DOCX)": 80.0,
     "PDF": 61.8,
     "Presentations (PPT)": 72.7
    }
   },
   "deepseek-v4-flash": {
    "scores": {
     "Documents (DOCX)": 85.5,
     "PDF": 73.6,
     "Presentations (PPT)": 64.8
    }
   },
   "deepseek-v4.1-flash-1017": {
    "scores": {
     "Documents (DOCX)": 81.3,
     "PDF": 89.2,
     "Presentations (PPT)": 80.4
    }
   },
   "deepseek-v4.1-flash": {
    "scores": {
     "Documents (DOCX)": 74.9,
     "PDF": 86.8,
     "Presentations (PPT)": 75.8
    }
   },
   "gemini-3.1-pro": {
    "scores": {
     "Documents (DOCX)": 49.2,
     "PDF": 59.9,
     "Presentations (PPT)": 67.9
    }
   },
   "gemini-3.7-flash": {
    "scores": {
     "Documents (DOCX)": 80.9,
     "PDF": 76.1,
     "Presentations (PPT)": 68.1
    }
   },
   "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
    }
   },
   "glm-5.3": {
    "scores": {
     "Documents (DOCX)": 82.8,
     "PDF": 75.5,
     "Presentations (PPT)": 76.1
    }
   },
   "gpt-5.5": {
    "scores": {
     "Documents (DOCX)": 67.5,
     "PDF": 66.5,
     "Presentations (PPT)": 74.0
    }
   },
   "gpt-5.6-luna": {
    "scores": {
     "Documents (DOCX)": 77.4,
     "PDF": 78.5,
     "Presentations (PPT)": 73.7
    }
   },
   "gpt-5.6-sol": {
    "scores": {
     "Documents (DOCX)": 85.2,
     "PDF": 78.0,
     "Presentations (PPT)": 82.4
    }
   },
   "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-sol": {
    "scores": {
     "Documents (DOCX)": 89.4,
     "PDF": 83.0,
     "Presentations (PPT)": 78.7
    }
   },
   "gpt-6.1-sol": {
    "scores": {
     "Documents (DOCX)": 86.2,
     "PDF": 79.1,
     "Presentations (PPT)": 82.3
    }
   },
   "grok-4.5": {
    "scores": {
     "Documents (DOCX)": 82.4,
     "PDF": 68.8,
     "Presentations (PPT)": 82.6
    }
   },
   "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": {
    "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
    }
   },
   "minimax-m3": {
    "scores": {
     "Documents (DOCX)": 73.5,
     "PDF": 65.4,
     "Presentations (PPT)": 63.8
    }
   },
   "mistral-large-4": {
    "scores": {
     "Documents (DOCX)": 81.1,
     "PDF": 52.1,
     "Presentations (PPT)": 68.1
    }
   },
   "muse-spark-1.2": {
    "scores": {
     "Documents (DOCX)": 83.9,
     "PDF": 71.5,
     "Presentations (PPT)": 57.1
    }
   },
   "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-lightning-3.5": {
    "scores": {
     "Documents (DOCX)": 42.8,
     "PDF": 44.5,
     "Presentations (PPT)": 44.0
    }
   },
   "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)": 75.2,
     "PDF": 57.8,
     "Presentations (PPT)": 56.7
    }
   },
   "nemotron-ultra-3": {
    "scores": {
     "Documents (DOCX)": 65.6,
     "PDF": 55.7,
     "Presentations (PPT)": 57.7
    }
   },
   "qwen3.8-27b": {
    "scores": {
     "Documents (DOCX)": 79.9,
     "PDF": 82.3,
     "Presentations (PPT)": 69.9
    }
   },
   "qwen3.8-flash": {
    "scores": {
     "Documents (DOCX)": 81.6,
     "PDF": 70.8,
     "Presentations (PPT)": 70.6
    }
   },
   "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
    }
   },
   "kimi-k3.1-pro": {
    "scores": {
     "Documents (DOCX)": 78.8,
     "PDF": 82.5,
     "Presentations (PPT)": 74.2
    }
   }
  }
 }
}