{
 "title": "SO-Arena: Benchmarking the Capability Spectrum and Reliability of System One Models",
 "families": [
  {
   "id": "semantic",
   "name": "Semantic understanding",
   "color": "#4b79a5"
  },
  {
   "id": "knowledge",
   "name": "Knowledge and reasoning",
   "color": "#5f9668"
  },
  {
   "id": "safety",
   "name": "Safety and policy compliance",
   "color": "#ad5563"
  },
  {
   "id": "evidence",
   "name": "Evidence assessment",
   "color": "#c0904a"
  },
  {
   "id": "agent",
   "name": "Agent task completion",
   "color": "#7a69a8"
  }
 ],
 "domains": [
  {
   "id": "sentiment",
   "name": "Sentiment & Emotion Understanding",
   "short": "Sentiment",
   "family": "semantic",
   "tasks": 200,
   "tiers": null,
   "leaderboard_tier": "All tasks",
   "leaderboard_tasks": 200,
   "formats": "choice, score",
   "sources": [
    "Amazon Reviews 2023",
    "GoEmotions",
    "SST-5",
    "DynaSent",
    "ARTS"
   ],
   "sources_by_tier": false
  },
  {
   "id": "intent",
   "name": "Intent & Routing",
   "short": "Intent",
   "family": "semantic",
   "tasks": 300,
   "tiers": {
    "E": 100,
    "M": 100,
    "H": 100
   },
   "leaderboard_tier": "Hard",
   "leaderboard_tasks": 100,
   "formats": "choice, noul",
   "sources": [
    "CLINC150, BANKING77, SGD",
    "NatCS",
    "NLU++"
   ],
   "sources_by_tier": true
  },
  {
   "id": "math",
   "name": "Math",
   "short": "Math",
   "family": "knowledge",
   "tasks": 200,
   "tiers": {
    "E": 50,
    "M": 50,
    "H": 70,
    "V": 30
   },
   "leaderboard_tier": "Hard + Very hard",
   "leaderboard_tasks": 100,
   "formats": "choice",
   "sources": [
    "GSM8K-Platinum",
    "MATH500",
    "AIME 2025-2026"
   ],
   "sources_by_tier": false
  },
  {
   "id": "code",
   "name": "Code Understanding",
   "short": "Code",
   "family": "knowledge",
   "tasks": 200,
   "tiers": {
    "E": 60,
    "M": 70,
    "H": 70
   },
   "leaderboard_tier": "Hard",
   "leaderboard_tasks": 70,
   "formats": "noul",
   "sources": [
    "CodeJudgeBench",
    "LiveCodeBench",
    "LeetCodeDataset"
   ],
   "sources_by_tier": false
  },
  {
   "id": "science",
   "name": "Scientific Reasoning",
   "short": "Science",
   "family": "knowledge",
   "tasks": 200,
   "tiers": {
    "E": 50,
    "M": 70,
    "H": 80
   },
   "leaderboard_tier": "Hard",
   "leaderboard_tasks": 80,
   "formats": "choice",
   "sources": [
    "ARC-Challenge",
    "MMLU-Redux",
    "GPQA Diamond"
   ],
   "sources_by_tier": true
  },
  {
   "id": "knowledge",
   "name": "Knowledge & Commonsense",
   "short": "Knowledge",
   "family": "knowledge",
   "tasks": 200,
   "tiers": null,
   "leaderboard_tier": "All tasks",
   "leaderboard_tasks": 200,
   "formats": "choice",
   "sources": [
    "MMLU-Redux",
    "HellaSwag",
    "CommonsenseQA"
   ],
   "sources_by_tier": false
  },
  {
   "id": "safety",
   "name": "Safety",
   "short": "Safety",
   "family": "safety",
   "tasks": 500,
   "tiers": {
    "E": 177,
    "M": 165,
    "H": 158
   },
   "leaderboard_tier": "Hard",
   "leaderboard_tasks": 158,
   "formats": "choice, noul",
   "sources": [
    "RAGTruth",
    "ReasoningShield",
    "HarmThoughts",
    "DynaBench",
    "Aegis 2.0",
    "OR-Bench"
   ],
   "sources_by_tier": false
  },
  {
   "id": "retrieval",
   "name": "Retrieval & Reranking",
   "short": "Retrieval",
   "family": "evidence",
   "tasks": 300,
   "tiers": {
    "E": 100,
    "M": 100,
    "H": 100
   },
   "leaderboard_tier": "Hard",
   "leaderboard_tasks": 100,
   "formats": "noul, score",
   "sources": [
    "TREC DL",
    "BRIGHT, Touché 2020",
    "ACORD"
   ],
   "sources_by_tier": true
  },
  {
   "id": "verification",
   "name": "Task Completion Verification",
   "short": "Verification",
   "family": "evidence",
   "tasks": 200,
   "tiers": {
    "E": 91,
    "M": 89,
    "H": 20
   },
   "leaderboard_tier": "Hard",
   "leaderboard_tasks": 20,
   "formats": "choice, noul",
   "sources": [
    "BrowseComp-Plus",
    "CALVIN",
    "τ-bench",
    "MiniGrid",
    "EAI-VirtualHome"
   ],
   "sources_by_tier": false
  },
  {
   "id": "tools",
   "name": "Tool & API Use",
   "short": "Tools",
   "family": "agent",
   "tasks": 282,
   "tiers": {
    "E": 100,
    "M": 78,
    "H": 104
   },
   "leaderboard_tier": "Hard",
   "leaderboard_tasks": 104,
   "formats": "live sequence",
   "sources": [
    "BFCL",
    "ToolSandbox",
    "BFCL"
   ],
   "sources_by_tier": true
  },
  {
   "id": "web",
   "name": "Text-based Web Interaction",
   "short": "Web",
   "family": "agent",
   "tasks": 337,
   "tiers": {
    "E": 100,
    "M": 137,
    "H": 100
   },
   "leaderboard_tier": "Hard",
   "leaderboard_tasks": 100,
   "formats": "choice, noul; sequence",
   "sources": [
    "Mind2Web",
    "BreakingWeb"
   ],
   "sources_by_tier": true
  }
 ],
 "models": [
  {
   "key": "GLiDE",
   "name": "GLiDE",
   "developer": "Fastino",
   "size": "27B",
   "group": "open",
   "reasoning": "No",
   "release": "fastino/GLiDE",
   "release_url": "https://huggingface.co/fastino/GLiDE",
   "logo": "assets/logos/glide.png",
   "score": 70.212,
   "domains": {
    "sentiment": 89.5,
    "intent": 75.6651,
    "math": 54.0,
    "code": 57.1429,
    "science": 55.0,
    "knowledge": 95.0,
    "safety": 74.6835,
    "retrieval": 57.0,
    "verification": 100.0,
    "tools": 65.3846,
    "web": 48.9559
   },
   "tiers": {
    "sentiment": {
     "All": 90
    },
    "intent": {
     "E": 100,
     "M": 84,
     "H": 76
    },
    "math": {
     "E": 98,
     "M": 86,
     "H": 64,
     "V": 30
    },
    "code": {
     "E": 93,
     "M": 69,
     "H": 57
    },
    "science": {
     "E": 96,
     "M": 87,
     "H": 55
    },
    "knowledge": {
     "All": 95
    },
    "safety": {
     "E": 87,
     "M": 84,
     "H": 75
    },
    "retrieval": {
     "E": 99,
     "M": 84,
     "H": 57
    },
    "verification": {
     "E": 100,
     "M": 99,
     "H": 100
    },
    "tools": {
     "E": 97,
     "M": 81,
     "H": 65
    },
    "web": {
     "E": 99,
     "M": 81,
     "H": 49
    }
   },
   "reliability": {
    "R": 77.57,
    "S": 0.9424,
    "N": 0.6384
   },
   "latency": {
    "mean_ms": 113.0,
    "timing": "one H100",
    "p90_ms": 214.0,
    "p99_ms": 774.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 83.44,
    "runs": [
     82.94,
     83.28,
     82.69,
     82.43,
     83.12,
     82.64,
     83.39,
     82.62,
     82.43,
     82.43,
     82.96,
     82.3
    ],
    "min": 82.3,
    "max": 83.44
   }
  },
  {
   "key": "KnowLine-31B-Gen1",
   "name": "KnowLine-31B",
   "developer": "PelaAI",
   "size": "31B",
   "group": "open",
   "reasoning": "No",
   "release": "PelaAI/KnowLine-31B-Gen1",
   "release_url": "https://huggingface.co/PelaAI/KnowLine-31B-Gen1",
   "logo": null,
   "score": 70.1908,
   "domains": {
    "sentiment": 88.25,
    "intent": 78.1869,
    "math": 62.0,
    "code": 67.1429,
    "science": 52.5,
    "knowledge": 93.0,
    "safety": 77.2152,
    "retrieval": 64.0,
    "verification": 90.0,
    "tools": 55.7692,
    "web": 44.0341
   },
   "tiers": {
    "sentiment": {
     "All": 88
    },
    "intent": {
     "E": 100,
     "M": 83,
     "H": 78
    },
    "math": {
     "E": 98,
     "M": 98,
     "H": 71,
     "V": 40
    },
    "code": {
     "E": 92,
     "M": 79,
     "H": 67
    },
    "science": {
     "E": 98,
     "M": 94,
     "H": 52
    },
    "knowledge": {
     "All": 93
    },
    "safety": {
     "E": 86,
     "M": 83,
     "H": 77
    },
    "retrieval": {
     "E": 99,
     "M": 89,
     "H": 64
    },
    "verification": {
     "E": 97,
     "M": 96,
     "H": 90
    },
    "tools": {
     "E": 97,
     "M": 84,
     "H": 56
    },
    "web": {
     "E": 100,
     "M": 74,
     "H": 44
    }
   },
   "reliability": {
    "R": 81.01,
    "S": 0.9237,
    "N": 0.7104
   },
   "latency": {
    "mean_ms": 177.0,
    "timing": "one H100",
    "p90_ms": 515.0,
    "p99_ms": 959.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 84.45,
    "runs": [
     84.03,
     83.49,
     84.66,
     84.56,
     85.2,
     84.45,
     84.77,
     84.45,
     85.84,
     85.09,
     85.73,
     85.09
    ],
    "min": 83.49,
    "max": 85.84
   }
  },
  {
   "key": "Sieve-31B-v3",
   "name": "Sieve-31B-v3",
   "developer": "Sthanika AI",
   "size": "31B",
   "group": "open",
   "reasoning": "No",
   "release": "sthanika-ai/Sieve-31B-v3",
   "release_url": "https://huggingface.co/sthanika-ai/Sieve-31B-v3",
   "logo": null,
   "score": 67.6503,
   "domains": {
    "sentiment": 88.0,
    "intent": 68.0214,
    "math": 60.0,
    "code": 65.7143,
    "science": 55.0,
    "knowledge": 93.0,
    "safety": 78.481,
    "retrieval": 63.5,
    "verification": 70.0,
    "tools": 59.6154,
    "web": 42.8211
   },
   "tiers": {
    "sentiment": {
     "All": 88
    },
    "intent": {
     "E": 99,
     "M": 81,
     "H": 68
    },
    "math": {
     "E": 100,
     "M": 94,
     "H": 66,
     "V": 47
    },
    "code": {
     "E": 93,
     "M": 77,
     "H": 66
    },
    "science": {
     "E": 98,
     "M": 93,
     "H": 55
    },
    "knowledge": {
     "All": 93
    },
    "safety": {
     "E": 86,
     "M": 82,
     "H": 78
    },
    "retrieval": {
     "E": 98,
     "M": 84,
     "H": 64
    },
    "verification": {
     "E": 97,
     "M": 98,
     "H": 70
    },
    "tools": {
     "E": 96,
     "M": 86,
     "H": 60
    },
    "web": {
     "E": 99,
     "M": 60,
     "H": 43
    }
   },
   "reliability": {
    "R": 81.31,
    "S": 0.9353,
    "N": 0.7069
   },
   "latency": {
    "mean_ms": 265.0,
    "timing": "one H100",
    "p90_ms": 668.0,
    "p99_ms": 2475.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 83.28,
    "runs": [
     83.92,
     83.71,
     84.66,
     83.71,
     84.03,
     84.13,
     83.6,
     83.92,
     85.41,
     84.03,
     85.09,
     84.35
    ],
    "min": 83.28,
    "max": 85.41
   }
  },
  {
   "key": "pplx-decider-27B",
   "name": "Decider-27B",
   "developer": "Perplexity",
   "size": "27B",
   "group": "open",
   "reasoning": "No",
   "release": "perplexity-ai/pplx-decider-v1.2",
   "release_url": "https://huggingface.co/perplexity-ai/pplx-decider-v1.2",
   "logo": "assets/logos/pplx-decider-27b.png",
   "score": 66.7044,
   "domains": {
    "sentiment": 88.75,
    "intent": 69.8437,
    "math": 53.0,
    "code": 61.4286,
    "science": 52.5,
    "knowledge": 95.0,
    "safety": 74.6835,
    "retrieval": 56.0,
    "verification": 65.0,
    "tools": 67.3077,
    "web": 50.2353
   },
   "tiers": {
    "sentiment": {
     "All": 89
    },
    "intent": {
     "E": 98,
     "M": 83,
     "H": 70
    },
    "math": {
     "E": 96,
     "M": 90,
     "H": 61,
     "V": 33
    },
    "code": {
     "E": 87,
     "M": 73,
     "H": 61
    },
    "science": {
     "E": 100,
     "M": 90,
     "H": 52
    },
    "knowledge": {
     "All": 95
    },
    "safety": {
     "E": 86,
     "M": 81,
     "H": 75
    },
    "retrieval": {
     "E": 99,
     "M": 88,
     "H": 56
    },
    "verification": {
     "E": 99,
     "M": 97,
     "H": 65
    },
    "tools": {
     "E": 97,
     "M": 84,
     "H": 67
    },
    "web": {
     "E": 99,
     "M": 79,
     "H": 50
    }
   },
   "reliability": {
    "R": 75.29,
    "S": 0.9242,
    "N": 0.6134
   },
   "latency": {
    "mean_ms": 153.0,
    "timing": "one H100",
    "p90_ms": 145.0,
    "p99_ms": 584.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 82.11,
    "runs": [
     81.95,
     82.8,
     82.75,
     82.11,
     82.75,
     81.52,
     82.11,
     82.43,
     82.43,
     83.17,
     82.27,
     82.06
    ],
    "min": 81.52,
    "max": 83.17
   }
  },
  {
   "key": "Kev-27B",
   "name": "Kev-27B",
   "developer": "J. Palmer",
   "size": "27B",
   "group": "open",
   "reasoning": "No",
   "release": "jaredpalmer/kev-27b",
   "release_url": "https://huggingface.co/jaredpalmer/kev-27b",
   "logo": null,
   "score": 61.3923,
   "domains": {
    "sentiment": 84.25,
    "intent": 68.1329,
    "math": 48.0,
    "code": 54.2857,
    "science": 53.75,
    "knowledge": 92.0,
    "safety": 63.9241,
    "retrieval": 59.0,
    "verification": 45.0,
    "tools": 61.5385,
    "web": 45.4341
   },
   "tiers": {
    "sentiment": {
     "All": 84
    },
    "intent": {
     "E": 97,
     "M": 82,
     "H": 68
    },
    "math": {
     "E": 100,
     "M": 78,
     "H": 56,
     "V": 30
    },
    "code": {
     "E": 87,
     "M": 71,
     "H": 54
    },
    "science": {
     "E": 98,
     "M": 87,
     "H": 54
    },
    "knowledge": {
     "All": 92
    },
    "safety": {
     "E": 79,
     "M": 73,
     "H": 64
    },
    "retrieval": {
     "E": 98,
     "M": 87,
     "H": 59
    },
    "verification": {
     "E": 89,
     "M": 89,
     "H": 45
    },
    "tools": {
     "E": 97,
     "M": 81,
     "H": 62
    },
    "web": {
     "E": 100,
     "M": 78,
     "H": 45
    }
   },
   "reliability": {
    "R": 80.78,
    "S": 0.8541,
    "N": 0.764
   },
   "latency": {
    "mean_ms": 90.3,
    "timing": "one H100",
    "p90_ms": 181.0,
    "p99_ms": 598.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 77.96,
    "runs": [
     77.0,
     78.06,
     77.74,
     77.21,
     77.64,
     77.0,
     76.68,
     77.96,
     76.78,
     77.21,
     76.89,
     77.1
    ],
    "min": 76.68,
    "max": 78.06
   }
  },
  {
   "key": "JevAny-27B",
   "name": "JevAny-27B",
   "developer": "SimpleJev",
   "size": "27B",
   "group": "open",
   "reasoning": "No",
   "release": "SimpleJev/JevAny-Qwen3.8-27B-LoRA",
   "release_url": "https://huggingface.co/SimpleJev/JevAny-Qwen3.8-27B-LoRA",
   "logo": "assets/logos/jevany-27b.png",
   "score": 58.3695,
   "domains": {
    "sentiment": 84.25,
    "intent": 52.1333,
    "math": 45.0,
    "code": 50.0,
    "science": 60.0,
    "knowledge": 94.5,
    "safety": 65.1899,
    "retrieval": 57.0,
    "verification": 35.0,
    "tools": 57.6923,
    "web": 41.2986
   },
   "tiers": {
    "sentiment": {
     "All": 84
    },
    "intent": {
     "E": 98,
     "M": 81,
     "H": 52
    },
    "math": {
     "E": 100,
     "M": 82,
     "H": 53,
     "V": 27
    },
    "code": {
     "E": 87,
     "M": 73,
     "H": 50
    },
    "science": {
     "E": 100,
     "M": 87,
     "H": 60
    },
    "knowledge": {
     "All": 94
    },
    "safety": {
     "E": 81,
     "M": 75,
     "H": 65
    },
    "retrieval": {
     "E": 92,
     "M": 70,
     "H": 57
    },
    "verification": {
     "E": 93,
     "M": 90,
     "H": 35
    },
    "tools": {
     "E": 98,
     "M": 84,
     "H": 58
    },
    "web": {
     "E": 99,
     "M": 71,
     "H": 41
    }
   },
   "reliability": {
    "R": 84.11,
    "S": 0.8917,
    "N": 0.7934
   },
   "latency": {
    "mean_ms": 93.9,
    "timing": "one H100",
    "p90_ms": 233.0,
    "p99_ms": 697.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 79.55,
    "runs": [
     79.13,
     80.4,
     79.87,
     80.09,
     79.77,
     80.09,
     79.77,
     78.91,
     79.66,
     79.77,
     80.3,
     80.09
    ],
    "min": 78.91,
    "max": 80.4
   }
  },
  {
   "key": "StartLux-4B-AutoThink",
   "name": "StartLux Auto",
   "developer": "StartLux",
   "size": "4B",
   "group": "auto",
   "reasoning": "Auto",
   "release": "startlux-models/StartLux-Decision-v1.1-4B-AutoThink",
   "release_url": "https://huggingface.co/startlux-models/StartLux-Decision-v1.1-4B-AutoThink",
   "logo": "assets/logos/startlux-auto.png",
   "score": 58.629,
   "domains": {
    "sentiment": 83.75,
    "intent": 69.8468,
    "math": 71.0,
    "code": 54.2857,
    "science": 51.25,
    "knowledge": 85.0,
    "safety": 62.0253,
    "retrieval": 56.0,
    "verification": 35.0,
    "tools": 46.1538,
    "web": 30.6075
   },
   "tiers": {
    "sentiment": {
     "All": 84
    },
    "intent": {
     "E": 96,
     "M": 71,
     "H": 70
    },
    "math": {
     "E": 98,
     "M": 90,
     "H": 81,
     "V": 47
    },
    "code": {
     "E": 80,
     "M": 63,
     "H": 54
    },
    "science": {
     "E": 90,
     "M": 86,
     "H": 51
    },
    "knowledge": {
     "All": 85
    },
    "safety": {
     "E": 74,
     "M": 66,
     "H": 62
    },
    "retrieval": {
     "E": 98,
     "M": 82,
     "H": 56
    },
    "verification": {
     "E": 88,
     "M": 76,
     "H": 35
    },
    "tools": {
     "E": 92,
     "M": 76,
     "H": 46
    },
    "web": {
     "E": 92,
     "M": 50,
     "H": 31
    }
   },
   "reliability": null,
   "latency": {
    "mean_ms": 3084.0,
    "timing": "one H100",
    "p90_ms": 10032.0,
    "p99_ms": 21870.0
   },
   "note": "Switches automatically between System One and System Two; not in the reliability set."
  },
  {
   "key": "StartLux-4B-Instant",
   "name": "StartLux Instant",
   "developer": "StartLux",
   "size": "4B",
   "group": "open",
   "reasoning": "No",
   "release": "startlux-models/StartLux-Decision-v1.1-4B-Instant",
   "release_url": "https://huggingface.co/startlux-models/StartLux-Decision-v1.1-4B-Instant",
   "logo": "assets/logos/startlux-instant.png",
   "score": 58.9563,
   "domains": {
    "sentiment": 87.75,
    "intent": 70.1976,
    "math": 42.0,
    "code": 55.7143,
    "science": 46.25,
    "knowledge": 88.5,
    "safety": 63.2911,
    "retrieval": 54.5,
    "verification": 55.0,
    "tools": 53.8462,
    "web": 31.4698
   },
   "tiers": {
    "sentiment": {
     "All": 88
    },
    "intent": {
     "E": 94,
     "M": 71,
     "H": 70
    },
    "math": {
     "E": 98,
     "M": 88,
     "H": 43,
     "V": 40
    },
    "code": {
     "E": 80,
     "M": 69,
     "H": 56
    },
    "science": {
     "E": 96,
     "M": 84,
     "H": 46
    },
    "knowledge": {
     "All": 88
    },
    "safety": {
     "E": 82,
     "M": 73,
     "H": 63
    },
    "retrieval": {
     "E": 99,
     "M": 82,
     "H": 54
    },
    "verification": {
     "E": 86,
     "M": 80,
     "H": 55
    },
    "tools": {
     "E": 96,
     "M": 79,
     "H": 54
    },
    "web": {
     "E": 98,
     "M": 39,
     "H": 31
    }
   },
   "reliability": {
    "R": 78.9,
    "S": 0.8904,
    "N": 0.6991
   },
   "latency": {
    "mean_ms": 35.1,
    "timing": "one H100",
    "p90_ms": 63.2,
    "p99_ms": 138.0
   },
   "note": "Null discrimination uses the model's own null output.",
   "order_acc": {
    "n": 939,
    "original": 76.84,
    "runs": [
     76.52,
     76.2,
     76.14,
     75.88,
     75.35,
     76.25,
     76.3,
     76.18,
     75.99,
     76.78,
     76.25,
     75.49
    ],
    "min": 75.35,
    "max": 76.84
   }
  },
  {
   "key": "Clef-27B",
   "name": "Clef-27B",
   "developer": "Cloudflare",
   "size": "27B",
   "group": "open",
   "reasoning": "No",
   "release": "Cloudflare/clef",
   "release_url": "https://huggingface.co/Cloudflare/clef",
   "logo": "assets/logos/clef-27b.png",
   "score": 56.8675,
   "domains": {
    "sentiment": 85.0,
    "intent": 63.9179,
    "math": 38.0,
    "code": 60.0,
    "science": 51.25,
    "knowledge": 94.5,
    "safety": 59.4937,
    "retrieval": 61.0,
    "verification": 15.0,
    "tools": 54.8077,
    "web": 42.573
   },
   "tiers": {
    "sentiment": {
     "All": 85
    },
    "intent": {
     "E": 97,
     "M": 79,
     "H": 64
    },
    "math": {
     "E": 96,
     "M": 74,
     "H": 47,
     "V": 17
    },
    "code": {
     "E": 92,
     "M": 70,
     "H": 60
    },
    "science": {
     "E": 98,
     "M": 91,
     "H": 51
    },
    "knowledge": {
     "All": 94
    },
    "safety": {
     "E": 78,
     "M": 75,
     "H": 59
    },
    "retrieval": {
     "E": 97,
     "M": 88,
     "H": 61
    },
    "verification": {
     "E": 88,
     "M": 83,
     "H": 15
    },
    "tools": {
     "E": 92,
     "M": 86,
     "H": 55
    },
    "web": {
     "E": 98,
     "M": 73,
     "H": 43
    }
   },
   "reliability": {
    "R": 73.21,
    "S": 0.8482,
    "N": 0.6319
   },
   "latency": {
    "mean_ms": 108.0,
    "timing": "one H100",
    "p90_ms": 184.0,
    "p99_ms": 533.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 75.51,
    "runs": [
     74.76,
     76.09,
     74.49,
     76.2,
     74.76,
     74.55,
     75.24,
     75.83,
     75.67,
     75.03,
     75.19,
     73.91
    ],
    "min": 73.91,
    "max": 76.2
   }
  },
  {
   "key": "H2O-Lightning-4B",
   "name": "H2O Lightning",
   "developer": "H2O.ai",
   "size": "4B",
   "group": "open",
   "reasoning": "No",
   "release": "h2oai/h2o-lightning-4b",
   "release_url": "https://huggingface.co/h2oai/h2o-lightning-4b",
   "logo": "assets/logos/h2o-lightning.png",
   "score": 54.8581,
   "domains": {
    "sentiment": 85.25,
    "intent": 63.0468,
    "math": 44.0,
    "code": 45.7143,
    "science": 43.75,
    "knowledge": 91.0,
    "safety": 63.9241,
    "retrieval": 61.0,
    "verification": 40.0,
    "tools": 37.5,
    "web": 28.2538
   },
   "tiers": {
    "sentiment": {
     "All": 85
    },
    "intent": {
     "E": 94,
     "M": 66,
     "H": 63
    },
    "math": {
     "E": 98,
     "M": 72,
     "H": 44,
     "V": 43
    },
    "code": {
     "E": 70,
     "M": 54,
     "H": 46
    },
    "science": {
     "E": 92,
     "M": 80,
     "H": 44
    },
    "knowledge": {
     "All": 91
    },
    "safety": {
     "E": 82,
     "M": 68,
     "H": 64
    },
    "retrieval": {
     "E": 98,
     "M": 87,
     "H": 61
    },
    "verification": {
     "E": 85,
     "M": 80,
     "H": 40
    },
    "tools": {
     "E": 96,
     "M": 66,
     "H": 38
    },
    "web": {
     "E": 94,
     "M": 36,
     "H": 28
    }
   },
   "reliability": {
    "R": 60.91,
    "S": 0.8897,
    "N": 0.417
   },
   "latency": {
    "mean_ms": 26.5,
    "timing": "one H100",
    "p90_ms": 48.4,
    "p99_ms": 155.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 74.23,
    "runs": [
     73.7,
     73.91,
     74.97,
     73.91,
     74.23,
     74.33,
     74.33,
     73.48,
     73.27,
     73.06,
     74.44,
     74.55
    ],
    "min": 73.06,
    "max": 74.97
   }
  },
  {
   "key": "Clef-flash-9B",
   "name": "Clef-Flash-9B",
   "developer": "Cloudflare",
   "size": "9B",
   "group": "open",
   "reasoning": "No",
   "release": "Cloudflare/clef-flash",
   "release_url": "https://huggingface.co/Cloudflare/clef-flash",
   "logo": "assets/logos/clef-flash-9b.png",
   "score": 51.6132,
   "domains": {
    "sentiment": 83.0,
    "intent": 64.3333,
    "math": 41.0,
    "code": 41.4286,
    "science": 47.5,
    "knowledge": 95.5,
    "safety": 45.5696,
    "retrieval": 57.5,
    "verification": 35.0,
    "tools": 31.7308,
    "web": 25.1827
   },
   "tiers": {
    "sentiment": {
     "All": 83
    },
    "intent": {
     "E": 99,
     "M": 42,
     "H": 64
    },
    "math": {
     "E": 94,
     "M": 68,
     "H": 51,
     "V": 17
    },
    "code": {
     "E": 70,
     "M": 59,
     "H": 41
    },
    "science": {
     "E": 98,
     "M": 94,
     "H": 48
    },
    "knowledge": {
     "All": 96
    },
    "safety": {
     "E": 70,
     "M": 66,
     "H": 46
    },
    "retrieval": {
     "E": 99,
     "M": 83,
     "H": 58
    },
    "verification": {
     "E": 76,
     "M": 58,
     "H": 35
    },
    "tools": {
     "E": 91,
     "M": 50,
     "H": 32
    },
    "web": {
     "E": 90,
     "M": 55,
     "H": 25
    }
   },
   "reliability": {
    "R": 71.0,
    "S": 0.8525,
    "N": 0.5913
   },
   "latency": {
    "mean_ms": 53.5,
    "timing": "one H100",
    "p90_ms": 69.7,
    "p99_ms": 230.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 72.36,
    "runs": [
     72.31,
     71.62,
     70.18,
     72.42,
     71.94,
     70.98,
     71.67,
     71.51,
     72.26,
     71.3,
     71.19,
     71.35
    ],
    "min": 70.18,
    "max": 72.42
   }
  },
  {
   "key": "pplx-decider-flash-4B",
   "name": "Decider Flash",
   "developer": "Perplexity",
   "size": "4B",
   "group": "open",
   "reasoning": "No",
   "release": "perplexity-ai/pplx-decider-flash-v1.2",
   "release_url": "https://huggingface.co/perplexity-ai/pplx-decider-flash-v1.2",
   "logo": "assets/logos/pplx-flash-4b.png",
   "score": 52.0001,
   "domains": {
    "sentiment": 88.0,
    "intent": 59.3889,
    "math": 34.0,
    "code": 52.8571,
    "science": 40.0,
    "knowledge": 91.0,
    "safety": 60.7595,
    "retrieval": 56.0,
    "verification": 15.0,
    "tools": 41.3462,
    "web": 33.6492
   },
   "tiers": {
    "sentiment": {
     "All": 88
    },
    "intent": {
     "E": 98,
     "M": 75,
     "H": 59
    },
    "math": {
     "E": 96,
     "M": 76,
     "H": 40,
     "V": 20
    },
    "code": {
     "E": 77,
     "M": 64,
     "H": 53
    },
    "science": {
     "E": 92,
     "M": 83,
     "H": 40
    },
    "knowledge": {
     "All": 91
    },
    "safety": {
     "E": 79,
     "M": 65,
     "H": 61
    },
    "retrieval": {
     "E": 99,
     "M": 80,
     "H": 56
    },
    "verification": {
     "E": 88,
     "M": 83,
     "H": 15
    },
    "tools": {
     "E": 95,
     "M": 68,
     "H": 41
    },
    "web": {
     "E": 94,
     "M": 56,
     "H": 34
    }
   },
   "reliability": {
    "R": 67.92,
    "S": 0.908,
    "N": 0.508
   },
   "latency": {
    "mean_ms": 39.8,
    "timing": "one H100",
    "p90_ms": 39.2,
    "p99_ms": 149.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 72.63,
    "runs": [
     72.31,
     73.75,
     72.36,
     72.2,
     72.74,
     71.88,
     71.51,
     71.73,
     73.0,
     71.67,
     72.04,
     72.95
    ],
    "min": 71.51,
    "max": 73.75
   }
  },
  {
   "key": "JevAny-4B",
   "name": "JevAny-4B",
   "developer": "SimpleJev",
   "size": "4B",
   "group": "open",
   "reasoning": "No",
   "release": "SimpleJev/JevAny-Qwen3.5-4B-LoRA",
   "release_url": "https://huggingface.co/SimpleJev/JevAny-Qwen3.5-4B-LoRA",
   "logo": "assets/logos/jevany-4b.png",
   "score": 50.9202,
   "domains": {
    "sentiment": 83.25,
    "intent": 60.8667,
    "math": 33.0,
    "code": 47.1429,
    "science": 43.75,
    "knowledge": 89.5,
    "safety": 45.5696,
    "retrieval": 52.0,
    "verification": 45.0,
    "tools": 41.3462,
    "web": 18.6973
   },
   "tiers": {
    "sentiment": {
     "All": 83
    },
    "intent": {
     "E": 97,
     "M": 67,
     "H": 61
    },
    "math": {
     "E": 92,
     "M": 60,
     "H": 39,
     "V": 20
    },
    "code": {
     "E": 73,
     "M": 66,
     "H": 47
    },
    "science": {
     "E": 96,
     "M": 84,
     "H": 44
    },
    "knowledge": {
     "All": 90
    },
    "safety": {
     "E": 69,
     "M": 63,
     "H": 46
    },
    "retrieval": {
     "E": 91,
     "M": 72,
     "H": 52
    },
    "verification": {
     "E": 71,
     "M": 63,
     "H": 45
    },
    "tools": {
     "E": 93,
     "M": 72,
     "H": 41
    },
    "web": {
     "E": 94,
     "M": 48,
     "H": 19
    }
   },
   "reliability": {
    "R": 70.54,
    "S": 0.8261,
    "N": 0.6024
   },
   "latency": {
    "mean_ms": 25.2,
    "timing": "one H100",
    "p90_ms": 56.2,
    "p99_ms": 173.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 68.8,
    "runs": [
     69.12,
     68.58,
     69.22,
     69.33,
     68.8,
     68.8,
     68.26,
     69.54,
     69.76,
     69.86,
     69.22,
     69.12
    ],
    "min": 68.26,
    "max": 69.86
   }
  },
  {
   "key": "SII-Decision-4B",
   "name": "SII-4B",
   "developer": "SII",
   "size": "4B",
   "group": "open",
   "reasoning": "No",
   "release": "jylin001206/SII-Decision-4B-LoRA",
   "release_url": "https://huggingface.co/jylin001206/SII-Decision-4B-LoRA",
   "logo": "assets/logos/sii-decision-4b.png",
   "score": 47.978,
   "domains": {
    "sentiment": 77.5,
    "intent": 40.3513,
    "math": 35.0,
    "code": 51.4286,
    "science": 45.0,
    "knowledge": 83.5,
    "safety": 44.9367,
    "retrieval": 54.5,
    "verification": 60.0,
    "tools": 20.1923,
    "web": 15.3491
   },
   "tiers": {
    "sentiment": {
     "All": 78
    },
    "intent": {
     "E": 91,
     "M": 64,
     "H": 40
    },
    "math": {
     "E": 62,
     "M": 42,
     "H": 39,
     "V": 27
    },
    "code": {
     "E": 70,
     "M": 59,
     "H": 51
    },
    "science": {
     "E": 90,
     "M": 81,
     "H": 45
    },
    "knowledge": {
     "All": 84
    },
    "safety": {
     "E": 71,
     "M": 60,
     "H": 45
    },
    "retrieval": {
     "E": 95,
     "M": 84,
     "H": 54
    },
    "verification": {
     "E": 82,
     "M": 71,
     "H": 60
    },
    "tools": {
     "E": 84,
     "M": 43,
     "H": 20
    },
    "web": {
     "E": 91,
     "M": 49,
     "H": 15
    }
   },
   "reliability": {
    "R": 36.17,
    "S": 0.7201,
    "N": 0.1817
   },
   "latency": {
    "mean_ms": 40.6,
    "timing": "one H100",
    "p90_ms": 48.5,
    "p99_ms": 157.0
   },
   "note": "Latency measured with flash-linear-attention added (SII-4B + fla).",
   "order_acc": {
    "n": 939,
    "original": 66.13,
    "runs": [
     65.65,
     66.12,
     64.52,
     65.67,
     66.27,
     65.19,
     66.28,
     64.94,
     65.85,
     65.76,
     63.69,
     65.9
    ],
    "min": 63.69,
    "max": 66.28
   }
  },
  {
   "key": "Kev-4B",
   "name": "Kev-4B",
   "developer": "J. Palmer",
   "size": "4B",
   "group": "open",
   "reasoning": "No",
   "release": "jaredpalmer/kev-4b",
   "release_url": "https://huggingface.co/jaredpalmer/kev-4b",
   "logo": null,
   "score": 46.0843,
   "domains": {
    "sentiment": 74.25,
    "intent": 49.8618,
    "math": 29.0,
    "code": 55.7143,
    "science": 41.25,
    "knowledge": 82.5,
    "safety": 32.2785,
    "retrieval": 54.0,
    "verification": 50.0,
    "tools": 22.1154,
    "web": 15.9575
   },
   "tiers": {
    "sentiment": {
     "All": 74
    },
    "intent": {
     "E": 92,
     "M": 64,
     "H": 50
    },
    "math": {
     "E": 90,
     "M": 58,
     "H": 33,
     "V": 20
    },
    "code": {
     "E": 67,
     "M": 66,
     "H": 56
    },
    "science": {
     "E": 88,
     "M": 74,
     "H": 41
    },
    "knowledge": {
     "All": 82
    },
    "safety": {
     "E": 66,
     "M": 58,
     "H": 32
    },
    "retrieval": {
     "E": 93,
     "M": 80,
     "H": 54
    },
    "verification": {
     "E": 76,
     "M": 72,
     "H": 50
    },
    "tools": {
     "E": 81,
     "M": 46,
     "H": 22
    },
    "web": {
     "E": 83,
     "M": 39,
     "H": 16
    }
   },
   "reliability": {
    "R": 60.34,
    "S": 0.7231,
    "N": 0.5036
   },
   "latency": {
    "mean_ms": 25.2,
    "timing": "one H100",
    "p90_ms": 61.0,
    "p99_ms": 157.0
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 63.68,
    "runs": [
     63.9,
     63.37,
     64.54,
     62.19,
     64.11,
     63.68,
     61.87,
     62.94,
     64.22,
     63.37,
     64.32,
     63.15
    ],
    "min": 61.87,
    "max": 64.54
   }
  },
  {
   "key": "Jeva-4B",
   "name": "Jeva-4B",
   "developer": "NUS",
   "size": "4B",
   "group": "open",
   "reasoning": "No",
   "release": "LIQIIIII/Jeva-4B",
   "release_url": "https://huggingface.co/LIQIIIII/Jeva-4B",
   "logo": "assets/logos/jeva-4b.png",
   "score": 51.5717,
   "domains": {
    "sentiment": 81.25,
    "intent": 61.571,
    "math": 44.0,
    "code": 50.0,
    "science": 60.0,
    "knowledge": 76.0,
    "safety": 50.6329,
    "retrieval": 64.0,
    "verification": 45.0,
    "tools": 17.3077,
    "web": 17.527
   },
   "tiers": {
    "sentiment": {
     "All": 81
    },
    "intent": {
     "E": 87,
     "M": 61,
     "H": 62
    },
    "math": {
     "E": 84,
     "M": 62,
     "H": 53,
     "V": 23
    },
    "code": {
     "E": 82,
     "M": 64,
     "H": 50
    },
    "science": {
     "E": 82,
     "M": 73,
     "H": 60
    },
    "knowledge": {
     "All": 76
    },
    "safety": {
     "E": 68,
     "M": 64,
     "H": 51
    },
    "retrieval": {
     "E": 96,
     "M": 85,
     "H": 64
    },
    "verification": {
     "E": 85,
     "M": 79,
     "H": 45
    },
    "tools": {
     "E": 73,
     "M": 32,
     "H": 17
    },
    "web": {
     "E": 85,
     "M": 61,
     "H": 18
    }
   },
   "reliability": {
    "R": 45.78,
    "S": 1.0,
    "N": 0.2096
   },
   "latency": {
    "mean_ms": 60.9,
    "timing": "one H100",
    "p90_ms": 72.9,
    "p99_ms": 142.0
   },
   "note": "Order stability is 1 by construction (option cache); latency with a warm option cache.",
   "order_acc": {
    "n": 939,
    "original": 48.78,
    "runs": [
     48.78,
     48.78,
     48.78,
     48.78,
     48.78,
     48.78,
     48.78,
     48.78,
     48.78,
     48.78,
     48.78,
     48.78
    ],
    "min": 48.78,
    "max": 48.78
   }
  },
  {
   "key": "Laya",
   "name": "Laya",
   "developer": "ConvAI Innovations",
   "size": "0.4B",
   "group": "open",
   "reasoning": "No",
   "release": "convaiinnovations/laya",
   "release_url": "https://huggingface.co/convaiinnovations/laya",
   "logo": "assets/logos/laya.png",
   "score": 30.888,
   "domains": {
    "sentiment": 65.0,
    "intent": 29.7202,
    "math": 22.0,
    "code": 51.4286,
    "science": 23.75,
    "knowledge": 43.0,
    "safety": 34.8101,
    "retrieval": 38.0,
    "verification": 30.0,
    "tools": 0.0,
    "web": 2.0595
   },
   "tiers": {
    "sentiment": {
     "All": 65
    },
    "intent": {
     "E": 82,
     "M": 7,
     "H": 30
    },
    "math": {
     "E": 42,
     "M": 42,
     "H": 23,
     "V": 20
    },
    "code": {
     "E": 58,
     "M": 51,
     "H": 51
    },
    "science": {
     "E": 26,
     "M": 29,
     "H": 24
    },
    "knowledge": {
     "All": 43
    },
    "safety": {
     "E": 49,
     "M": 39,
     "H": 35
    },
    "retrieval": {
     "E": 86,
     "M": 62,
     "H": 38
    },
    "verification": {
     "E": 52,
     "M": 44,
     "H": 30
    },
    "tools": {
     "E": 1,
     "M": 1,
     "H": 0
    },
    "web": {
     "E": 44,
     "M": 6,
     "H": 2
    }
   },
   "reliability": {
    "R": 7.02,
    "S": 0.354,
    "N": 0.0139
   },
   "latency": {
    "mean_ms": 6.0,
    "timing": "one H100",
    "p90_ms": 10.2,
    "p99_ms": 37.2
   },
   "note": "Every option is read in full, without the package's default cut at 48 tokens.",
   "order_acc": {
    "n": 939,
    "original": 36.21,
    "runs": [
     34.03,
     32.69,
     32.64,
     34.45,
     33.76,
     32.85,
     34.19,
     35.14,
     33.01,
     34.56,
     32.64,
     32.8
    ],
    "min": 32.64,
    "max": 36.21
   }
  },
  {
   "key": "GPT-6 Sol Responses · medium",
   "name": "GPT-6 Sol",
   "developer": "OpenAI",
   "size": null,
   "group": "s2",
   "reasoning": "Medium effort",
   "release": "gpt-6-sol",
   "release_url": null,
   "logo": "assets/logos/gpt-6-sol.png",
   "score": 82.3207,
   "domains": {
    "sentiment": 87.75,
    "intent": 82.2861,
    "math": 100.0,
    "code": 92.8571,
    "science": 91.25,
    "knowledge": 94.0,
    "safety": 75.3165,
    "retrieval": 62.5,
    "verification": 100.0,
    "tools": 66.3462,
    "web": 53.2215
   },
   "tiers": {
    "sentiment": {
     "All": 88
    },
    "intent": {
     "E": 100,
     "M": 78,
     "H": 82
    },
    "math": {
     "E": 100,
     "M": 100,
     "H": 100,
     "V": 100
    },
    "code": {
     "E": 98,
     "M": 96,
     "H": 93
    },
    "science": {
     "E": 98,
     "M": 96,
     "H": 91
    },
    "knowledge": {
     "All": 94
    },
    "safety": {
     "E": 84,
     "M": 82,
     "H": 75
    },
    "retrieval": {
     "E": 99,
     "M": 91,
     "H": 62
    },
    "verification": {
     "E": 100,
     "M": 99,
     "H": 100
    },
    "tools": {
     "E": 99,
     "M": 88,
     "H": 66
    },
    "web": {
     "E": 98,
     "M": 76,
     "H": 53
    }
   },
   "reliability": null,
   "latency": {
    "mean_ms": 2835.9,
    "timing": "end-to-end API"
   },
   "note": null
  },
  {
   "key": "GPT-6 Luna Responses · medium",
   "name": "GPT-6 Luna",
   "developer": "OpenAI",
   "size": null,
   "group": "s2",
   "reasoning": "Medium effort",
   "release": "gpt-6-luna",
   "release_url": null,
   "logo": "assets/logos/gpt-6-luna.png",
   "score": 79.6796,
   "domains": {
    "sentiment": 84.75,
    "intent": 81.4849,
    "math": 99.0,
    "code": 85.7143,
    "science": 87.5,
    "knowledge": 90.0,
    "safety": 70.2532,
    "retrieval": 62.0,
    "verification": 100.0,
    "tools": 68.2692,
    "web": 47.5039
   },
   "tiers": {
    "sentiment": {
     "All": 85
    },
    "intent": {
     "E": 99,
     "M": 77,
     "H": 81
    },
    "math": {
     "E": 100,
     "M": 100,
     "H": 100,
     "V": 97
    },
    "code": {
     "E": 98,
     "M": 91,
     "H": 86
    },
    "science": {
     "E": 98,
     "M": 96,
     "H": 88
    },
    "knowledge": {
     "All": 90
    },
    "safety": {
     "E": 82,
     "M": 80,
     "H": 70
    },
    "retrieval": {
     "E": 98,
     "M": 91,
     "H": 62
    },
    "verification": {
     "E": 100,
     "M": 98,
     "H": 100
    },
    "tools": {
     "E": 96,
     "M": 86,
     "H": 68
    },
    "web": {
     "E": 99,
     "M": 77,
     "H": 48
    }
   },
   "reliability": null,
   "latency": {
    "mean_ms": 2697.9,
    "timing": "end-to-end API"
   },
   "note": null
  },
  {
   "key": "Jev 1.14 API",
   "name": "Jev API 1.14",
   "developer": "TypeSafe",
   "size": null,
   "group": "api",
   "reasoning": "Not disclosed",
   "release": "jev-1.14.0",
   "release_url": null,
   "logo": "assets/logos/jev-api.png",
   "score": 60.4287,
   "domains": {
    "sentiment": 85.5,
    "intent": 56.0934,
    "math": 53.0,
    "code": 58.5714,
    "science": 71.25,
    "knowledge": 92.5,
    "safety": 68.9873,
    "retrieval": 59.5,
    "verification": 20.0,
    "tools": 58.6538,
    "web": 40.6598
   },
   "tiers": {
    "sentiment": {
     "All": 86
    },
    "intent": {
     "E": 99,
     "M": 78,
     "H": 56
    },
    "math": {
     "E": 100,
     "M": 84,
     "H": 63,
     "V": 30
    },
    "code": {
     "E": 83,
     "M": 70,
     "H": 59
    },
    "science": {
     "E": 96,
     "M": 94,
     "H": 71
    },
    "knowledge": {
     "All": 92
    },
    "safety": {
     "E": 83,
     "M": 73,
     "H": 69
    },
    "retrieval": {
     "E": 99,
     "M": 83,
     "H": 60
    },
    "verification": {
     "E": 88,
     "M": 88,
     "H": 20
    },
    "tools": {
     "E": 97,
     "M": 75,
     "H": 59
    },
    "web": {
     "E": 99,
     "M": 73,
     "H": 41
    }
   },
   "reliability": {
    "R": 83.45,
    "S": 0.8936,
    "N": 0.7794
   },
   "latency": {
    "mean_ms": 248.6,
    "timing": "end-to-end API"
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 80.99,
    "runs": [
     79.29,
     81.1,
     80.14,
     79.87,
     80.6,
     81.15,
     81.52,
     80.88,
     80.72,
     79.71,
     81.24,
     79.93
    ],
    "min": 79.29,
    "max": 81.52
   }
  },
  {
   "key": "OpenAI Decisions",
   "name": "OpenAI Decisions",
   "developer": "OpenAI",
   "size": null,
   "group": "api",
   "reasoning": "Not disclosed",
   "release": "gpt-6-luna",
   "release_url": null,
   "logo": "assets/logos/openai-decisions.png",
   "score": 53.5854,
   "domains": {
    "sentiment": 73.75,
    "intent": 57.5199,
    "math": 33.0,
    "code": 65.7143,
    "science": 46.25,
    "knowledge": 86.0,
    "safety": 63.2911,
    "retrieval": 55.0,
    "verification": 40.0,
    "tools": 37.5,
    "web": 31.4139
   },
   "tiers": {
    "sentiment": {
     "All": 74
    },
    "intent": {
     "E": 99,
     "M": 77,
     "H": 58
    },
    "math": {
     "E": 74,
     "M": 46,
     "H": 34,
     "V": 30
    },
    "code": {
     "E": 88,
     "M": 71,
     "H": 66
    },
    "science": {
     "E": 94,
     "M": 81,
     "H": 46
    },
    "knowledge": {
     "All": 86
    },
    "safety": {
     "E": 71,
     "M": 67,
     "H": 63
    },
    "retrieval": {
     "E": 98,
     "M": 84,
     "H": 55
    },
    "verification": {
     "E": 79,
     "M": 75,
     "H": 40
    },
    "tools": {
     "E": 86,
     "M": 61,
     "H": 38
    },
    "web": {
     "E": 98,
     "M": 68,
     "H": 31
    }
   },
   "reliability": {
    "R": 61.05,
    "S": 0.6777,
    "N": 0.55
   },
   "latency": {
    "mean_ms": 301.7,
    "timing": "end-to-end API"
   },
   "note": null,
   "order_acc": {
    "n": 939,
    "original": 65.37,
    "runs": [
     66.01,
     66.99,
     66.35,
     67.57,
     67.41,
     64.67,
     67.36,
     66.19,
     65.55,
     66.24,
     66.56,
     65.5
    ],
    "min": 64.67,
    "max": 67.57
   }
  },
  {
   "key": "Gemini Decision 1",
   "name": "Gemini Decision 1",
   "developer": "Google",
   "size": null,
   "group": "api",
   "reasoning": "Not disclosed",
   "release": "gemini-decision-1",
   "release_url": null,
   "logo": "assets/logos/gemini-decision.png",
   "score": 61.2497,
   "domains": {
    "sentiment": 82.25,
    "intent": 49.5242,
    "math": 58.0,
    "code": 72.8571,
    "science": 61.25,
    "knowledge": 92.0,
    "safety": 65.1899,
    "retrieval": 63.0,
    "verification": 30.0,
    "tools": 59.6154,
    "web": 40.0596
   },
   "tiers": {
    "sentiment": {
     "All": 82
    },
    "intent": {
     "E": 99,
     "M": 80,
     "H": 50
    },
    "math": {
     "E": 98,
     "M": 94,
     "H": 66,
     "V": 40
    },
    "code": {
     "E": 92,
     "M": 70,
     "H": 73
    },
    "science": {
     "E": 96,
     "M": 91,
     "H": 61
    },
    "knowledge": {
     "All": 92
    },
    "safety": {
     "E": 83,
     "M": 75,
     "H": 65
    },
    "retrieval": {
     "E": 97,
     "M": 76,
     "H": 63
    },
    "verification": {
     "E": 90,
     "M": 89,
     "H": 30
    },
    "tools": {
     "E": 94,
     "M": 84,
     "H": 60
    },
    "web": {
     "E": 100,
     "M": 77,
     "H": 40
    }
   },
   "reliability": null,
   "latency": null,
   "note": "Added after the paper (2026-11-13): run on the paper-version release with the released evaluator; reliability and latency not measured.",
   "added": "2026-11-13"
  }
 ],
 "analysis": [
  {
   "domain": "knowledge",
   "category": "direct",
   "tasks": 200,
   "gap_glide": 5.0,
   "gap_glide_ci": [
    1.5,
    9.0
   ],
   "gap_startlux": -1.5,
   "gap_startlux_ci": [
    -6.0,
    3.0
   ],
   "backbone_gain": 4.38,
   "thinking_never": 85.0,
   "thinking_always": 88.0,
   "thinking_gain": 3.0
  },
  {
   "domain": "sentiment",
   "category": "direct",
   "tasks": 200,
   "gap_glide": 4.75,
   "gap_glide_ci": [
    1.2,
    8.5
   ],
   "gap_startlux": 3.0,
   "gap_startlux_ci": [
    -2.0,
    8.0
   ],
   "backbone_gain": 3.44,
   "thinking_never": 88.0,
   "thinking_always": 85.8,
   "thinking_gain": -2.2
  },
  {
   "domain": "retrieval",
   "category": "direct",
   "tasks": 200,
   "gap_glide": -5.75,
   "gap_glide_ci": [
    -11.2,
    -0.2
   ],
   "gap_startlux": -8.25,
   "gap_startlux_ci": [
    -14.5,
    -1.8
   ],
   "backbone_gain": 3.88,
   "thinking_never": 69.0,
   "thinking_always": 70.8,
   "thinking_gain": 1.8
  },
  {
   "domain": "intent",
   "category": "grounded",
   "tasks": 200,
   "gap_glide": 0.13,
   "gap_glide_ci": [
    -2.4,
    2.9
   ],
   "gap_startlux": -8.89,
   "gap_startlux_ci": [
    -12.0,
    -5.8
   ],
   "backbone_gain": 12.05,
   "thinking_never": 68.8,
   "thinking_always": 70.8,
   "thinking_gain": 2.0
  },
  {
   "domain": "safety",
   "category": "grounded",
   "tasks": 323,
   "gap_glide": 4.02,
   "gap_glide_ci": [
    -1.2,
    9.0
   ],
   "gap_startlux": -7.12,
   "gap_startlux_ci": [
    -12.7,
    -1.5
   ],
   "backbone_gain": 16.18,
   "thinking_never": 60.4,
   "thinking_always": 63.8,
   "thinking_gain": 3.4
  },
  {
   "domain": "verification",
   "category": "grounded",
   "tasks": 109,
   "gap_glide": 0.92,
   "gap_glide_ci": [
    0.0,
    2.8
   ],
   "gap_startlux": -22.94,
   "gap_startlux_ci": [
    -31.2,
    -15.6
   ],
   "backbone_gain": 17.43,
   "thinking_never": 58.7,
   "thinking_always": 66.1,
   "thinking_gain": 7.3
  },
  {
   "domain": "tools",
   "category": "grounded",
   "tasks": 182,
   "gap_glide": -3.59,
   "gap_glide_ci": [
    -9.1,
    1.8
   ],
   "gap_startlux": -11.19,
   "gap_startlux_ci": [
    -17.1,
    -5.2
   ],
   "backbone_gain": 25.56,
   "thinking_never": 61.9,
   "thinking_always": 64.5,
   "thinking_gain": 2.6
  },
  {
   "domain": "web",
   "category": "grounded",
   "tasks": 237,
   "gap_glide": 3.07,
   "gap_glide_ci": [
    -0.4,
    6.5
   ],
   "gap_startlux": -28.85,
   "gap_startlux_ci": [
    -33.8,
    -23.8
   ],
   "backbone_gain": 23.92,
   "thinking_never": 40.8,
   "thinking_always": 42.6,
   "thinking_gain": 1.8
  },
  {
   "domain": "math",
   "category": "derivation",
   "tasks": 120,
   "gap_glide": -26.67,
   "gap_glide_ci": [
    -34.2,
    -19.2
   ],
   "gap_startlux": -38.33,
   "gap_startlux_ci": [
    -47.5,
    -30.0
   ],
   "backbone_gain": 14.37,
   "thinking_never": 50.8,
   "thinking_always": 92.5,
   "thinking_gain": 41.7
  },
  {
   "domain": "science",
   "category": "derivation",
   "tasks": 150,
   "gap_glide": -21.33,
   "gap_glide_ci": [
    -28.7,
    -14.7
   ],
   "gap_startlux": -27.33,
   "gap_startlux_ci": [
    -35.3,
    -19.3
   ],
   "backbone_gain": 8.33,
   "thinking_never": 58.7,
   "thinking_always": 70.7,
   "thinking_gain": 12.0
  },
  {
   "domain": "code",
   "category": "derivation",
   "tasks": 140,
   "gap_glide": -25.71,
   "gap_glide_ci": [
    -34.285714285714285,
    -17.142857142857142
   ],
   "gap_startlux": -26.43,
   "gap_startlux_ci": [
    -35.714285714285715,
    -17.142857142857142
   ],
   "backbone_gain": 7.68,
   "thinking_never": 55.7,
   "thinking_always": 61.4,
   "thinking_gain": 5.7
  }
 ],
 "jeva_cold_cache_ms": {
  "mean": 79.0,
  "p90": 88.2,
  "p99": 180.0
 },
 "totals": {
  "tasks": 2919,
  "static": 2300,
  "tools": 282,
  "web": 337,
  "leaderboard_tasks": 1232,
  "domains": 11,
  "families": 5,
  "models": 21,
  "added_models": 1
 },
 "provenance": {
  "ranking": {
   "policy": "highest-domain-math-hard-plus-ungraded-v2",
   "sha256": "c42a45117d9ce81bf0bd244a96e453441b37d4ca0f5c6be49e96251ea5cbbcc3"
  },
  "paper_overleaf_commit": null
 }
}
