[
  {
    "metric": "gpqa_diamond_pct",
    "label": "Kimi K3",
    "value": 93.1,
    "unit": "pct",
    "period": "2026",
    "geography": "Moonshot",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Grok 4",
    "value": 87.0,
    "unit": "pct",
    "period": "2026",
    "geography": "xAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "GPT-5",
    "value": 86.2,
    "unit": "pct",
    "period": "2026",
    "geography": "OpenAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Claude 3.7 Sonnet",
    "value": 79.7,
    "unit": "pct",
    "period": "2026",
    "geography": "Anthropic",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Grok 3",
    "value": 75.8,
    "unit": "pct",
    "period": "2026",
    "geography": "xAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Qwen3-Max",
    "value": 72.6,
    "unit": "pct",
    "period": "2026",
    "geography": "Alibaba",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "GPT-4.5",
    "value": 68.7,
    "unit": "pct",
    "period": "2026",
    "geography": "OpenAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Gemini 1.5 Pro",
    "value": 57.2,
    "unit": "pct",
    "period": "2026",
    "geography": "Google DeepMind",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Claude 3.5 Sonnet",
    "value": 54.0,
    "unit": "pct",
    "period": "2026",
    "geography": "Anthropic",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Grok-2",
    "value": 53.8,
    "unit": "pct",
    "period": "2026",
    "geography": "xAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Llama 3.1-405B",
    "value": 50.9,
    "unit": "pct",
    "period": "2026",
    "geography": "Meta AI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "GPT-4o",
    "value": 49.2,
    "unit": "pct",
    "period": "2026",
    "geography": "OpenAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Claude 3 Opus",
    "value": 47.2,
    "unit": "pct",
    "period": "2026",
    "geography": "Anthropic",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "GPT-4 Turbo (Apr 2024)",
    "value": 46.6,
    "unit": "pct",
    "period": "2026",
    "geography": "OpenAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "GPT-4 Turbo (Nov 2023)",
    "value": 42.4,
    "unit": "pct",
    "period": "2026",
    "geography": "OpenAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "GPT-4 (Mar 2023)",
    "value": 35.7,
    "unit": "pct",
    "period": "2026",
    "geography": "OpenAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "Claude 2",
    "value": 34.7,
    "unit": "pct",
    "period": "2026",
    "geography": "Anthropic",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  },
  {
    "metric": "gpqa_diamond_pct",
    "label": "GPT-4 (Jun 2023)",
    "value": 30.7,
    "unit": "pct",
    "period": "2026",
    "geography": "OpenAI",
    "source": "Epoch AI Benchmarking Hub",
    "source_url": "https://epoch.ai/benchmarks",
    "confidence": "High"
  }
]