{
  "name": "MultipleChat AI Benchmark Q2 2026",
  "canonical_url": "https://multiple.chat/ai-benchmark/q2-2026",
  "published": "2026-05-22",
  "license": "CC-BY-4.0",
  "methodology_note": "Public aggregate benchmark dataset used by the Q2 2026 benchmark page. Values are aggregate, anonymized comparison metrics.",
  "models": [
    {
      "model": "ChatGPT (GPT-5.5)",
      "overall_win_rate": 28.7,
      "comparisons": 50142,
      "avg_latency_seconds": 2.4,
      "cost_per_1000_wins_usd": 9.8,
      "hallucination_flag_rate": 4.2
    },
    {
      "model": "Claude Opus 4.6",
      "overall_win_rate": 27.9,
      "comparisons": 50142,
      "avg_latency_seconds": 3.1,
      "cost_per_1000_wins_usd": 11.4,
      "hallucination_flag_rate": 5.1
    },
    {
      "model": "Grok 4.5",
      "overall_win_rate": 21.9,
      "comparisons": 50142,
      "avg_latency_seconds": 1.9,
      "cost_per_1000_wins_usd": 5.9,
      "hallucination_flag_rate": 9.4
    },
    {
      "model": "Gemini 3.1 Pro",
      "overall_win_rate": 20.6,
      "comparisons": 50142,
      "avg_latency_seconds": 2.8,
      "cost_per_1000_wins_usd": 8.2,
      "hallucination_flag_rate": 6.7
    },
    {
      "model": "Perplexity Sonar",
      "overall_win_rate": 0.9,
      "comparisons": 12408,
      "avg_latency_seconds": 3.6,
      "cost_per_1000_wins_usd": 14.1,
      "hallucination_flag_rate": 3.1,
      "note": "Perplexity Sonar competes primarily on research and factual queries."
    }
  ]
}