{
  "generated": "2026-09-08",
  "benchmark": "ifeval",
  "results": [
    {
      "system": "GPT-4",
      "developer": "OpenAI",
      "value": 76.89,
      "date": "2023-11-14",
      "source": {
        "url": "https://arxiv.org/abs/2311.07911",
        "kind": "paper",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "split": "prompt-level strict",
        "notes": "Table 3 of the IFEval paper (responses collected November 2023); instruction-level strict 83.57, loose 79.30/85.37."
      }
    },
    {
      "system": "Llama 3.1 405B Instruct",
      "developer": "Meta",
      "value": 88.6,
      "date": "2024-07-23",
      "source": {
        "url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct",
        "kind": "developer-report",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "notes": "Llama 3.1 model card benchmark table (70B: 87.5, 8B: 80.4); Meta does not state which IFEval variant. Date is the Llama 3.1 release."
      }
    },
    {
      "system": "Qwen3-235B-A22B-Instruct-2507",
      "developer": "Alibaba",
      "value": 88.7,
      "date": "2025-07-21",
      "source": {
        "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507",
        "kind": "developer-report",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "notes": "Model card 'Alignment' table; the same table lists Kimi K2 at 89.8, Claude Opus 4 (non-thinking) 87.4 and GPT-4o-0327 83.9 as run by Qwen. Variant not stated."
      }
    },
    {
      "system": "Qwen3.5-27B",
      "developer": "Alibaba",
      "value": 95.0,
      "date": "2026-02-24",
      "source": {
        "url": "https://huggingface.co/Qwen/Qwen3.5-27B",
        "kind": "developer-report",
        "accessed": "2026-09-04"
      },
      "conditions": {
        "notes": "Model card benchmark table; same table lists GPT-5-mini (2025-08-07) at 93.9 as run by Qwen. Variant not stated. Also the top row on benchlm.ai's IFEval page."
      }
    }
  ]
}