{
  "timestamp": "2026-07-14T11:28:21.631070",
  "models": {
    "qwen3:0.6b": {
      "mmlu_pro": [
        {
          "q": "Which of the following is a primary function of mi",
          "correct": true,
          "latency_ms": 6780.056953430176
        },
        {
          "q": "What is the capital of Australia?",
          "correct": true,
          "latency_ms": 872.0991611480713
        },
        {
          "q": "The derivative of x^2 is:",
          "correct": false,
          "latency_ms": 2103.480100631714
        },
        {
          "q": "Which element has the atomic number 6?",
          "correct": true,
          "latency_ms": 1567.795991897583
        },
        {
          "q": "Photosynthesis occurs primarily in:",
          "correct": true,
          "latency_ms": 1507.2128772735596
        },
        {
          "q": "The square root of 144 is:",
          "correct": false,
          "latency_ms": 2097.5821018218994
        },
        {
          "q": "Who wrote 'Romeo and Juliet'?",
          "correct": true,
          "latency_ms": 1448.0640888214111
        },
        {
          "q": "The chemical symbol for gold is:",
          "correct": false,
          "latency_ms": 2000.0379085540771
        },
        {
          "q": "Speed of light in vacuum (m/s, approx):",
          "correct": true,
          "latency_ms": 2015.6428813934326
        },
        {
          "q": "Which planet is closest to the Sun?",
          "correct": true,
          "latency_ms": 1366.713047027588
        }
      ],
      "gsm8k": [
        {
          "q": "Janet has 3 apples. She buys 5 more. She gives 2 t",
          "correct": true,
          "latency_ms": 1880.690097808838
        },
        {
          "q": "A train travels 60 miles in 1 hour. At the same sp",
          "correct": false,
          "latency_ms": 2398.275852203369
        },
        {
          "q": "If a shirt costs $20 and is 25% off, what is the n",
          "correct": false,
          "latency_ms": 2641.2672996520996
        },
        {
          "q": "Tom is twice as old as Sam. Sam is 5. How old is T",
          "correct": false,
          "latency_ms": 2540.7559871673584
        },
        {
          "q": "A rectangle has length 8 and width 5. What is the ",
          "correct": false,
          "latency_ms": 2621.8550205230713
        },
        {
          "q": "Sarah reads 15 pages per day. How many pages in 4 ",
          "correct": false,
          "latency_ms": 2658.322811126709
        },
        {
          "q": "If 3x = 12, what is x?",
          "correct": false,
          "latency_ms": 2710.3610038757324
        },
        {
          "q": "A pizza is cut into 8 slices. If 3 are eaten, what",
          "correct": false,
          "latency_ms": 3081.5842151641846
        }
      ],
      "humaneval": [
        {
          "q": "Write a Python function `add(a, b)` that returns a",
          "correct": false,
          "latency_ms": 3155.5230617523193
        },
        {
          "q": "Write a Python function `factorial(n)` that return",
          "correct": false,
          "latency_ms": 2843.1859016418457
        },
        {
          "q": "Write a Python function `is_prime(n)` that returns",
          "correct": false,
          "latency_ms": 2621.018171310425
        },
        {
          "q": "Write a Python function `reverse_string(s)` that r",
          "correct": false,
          "latency_ms": 2659.5308780670166
        },
        {
          "q": "Write a Python function `fibonacci(n)` that return",
          "correct": false,
          "latency_ms": 2862.377882003784
        }
      ],
      "ifeval": [
        {
          "q": "List exactly 3 fruits, one per line, no extra text",
          "correct": true,
          "latency_ms": 1469.6540832519531
        },
        {
          "q": "Respond with ONLY the number 42, nothing else.",
          "correct": true,
          "latency_ms": 1254.9517154693604
        },
        {
          "q": "Write a sentence that contains the word 'sovereign",
          "correct": true,
          "latency_ms": 2199.057102203369
        },
        {
          "q": "Output exactly 5 words, all lowercase, describing ",
          "correct": false,
          "latency_ms": 2731.113910675049
        }
      ],
      "latencies": [
        6780.056953430176,
        872.0991611480713,
        2103.480100631714,
        1567.795991897583,
        1507.2128772735596,
        2097.5821018218994,
        1448.0640888214111,
        2000.0379085540771,
        2015.6428813934326,
        1366.713047027588,
        1880.690097808838,
        2398.275852203369,
        2641.2672996520996,
        2540.7559871673584,
        2621.8550205230713,
        2658.322811126709,
        2710.3610038757324,
        3081.5842151641846,
        3155.5230617523193,
        2843.1859016418457,
        2621.018171310425,
        2659.5308780670166,
        2862.377882003784,
        1469.6540832519531,
        1254.9517154693604,
        2199.057102203369,
        2731.113910675049
      ],
      "summary": {
        "mmlu_pro_pct": 70.0,
        "gsm8k_pct": 12.5,
        "humaneval_pct": 0.0,
        "ifeval_pct": 75.0,
        "composite_pct": 40.74074074074074,
        "median_latency_ms": 2398.275852203369,
        "p95_latency_ms": 3155.5230617523193
      }
    },
    "qwen3:1.7b": {
      "mmlu_pro": [
        {
          "q": "Which of the following is a primary function of mi",
          "correct": false,
          "latency_ms": 11216.080904006958
        },
        {
          "q": "What is the capital of Australia?",
          "correct": true,
          "latency_ms": 4791.441917419434
        },
        {
          "q": "The derivative of x^2 is:",
          "correct": false,
          "latency_ms": 7696.846723556519
        },
        {
          "q": "Which element has the atomic number 6?",
          "correct": true,
          "latency_ms": 6848.768949508667
        },
        {
          "q": "Photosynthesis occurs primarily in:",
          "correct": true,
          "latency_ms": 5726.844787597656
        },
        {
          "q": "The square root of 144 is:",
          "correct": false,
          "latency_ms": 8844.170808792114
        },
        {
          "q": "Who wrote 'Romeo and Juliet'?",
          "correct": true,
          "latency_ms": 8267.307043075562
        },
        {
          "q": "The chemical symbol for gold is:",
          "correct": false,
          "latency_ms": 8229.561805725098
        },
        {
          "q": "Speed of light in vacuum (m/s, approx):",
          "correct": false,
          "latency_ms": 9811.07211112976
        },
        {
          "q": "Which planet is closest to the Sun?",
          "correct": true,
          "latency_ms": 8163.851976394653
        }
      ],
      "gsm8k": [
        {
          "q": "Janet has 3 apples. She buys 5 more. She gives 2 t",
          "correct": true,
          "latency_ms": 8787.242889404297
        },
        {
          "q": "A train travels 60 miles in 1 hour. At the same sp",
          "correct": false,
          "latency_ms": 8994.663953781128
        },
        {
          "q": "If a shirt costs $20 and is 25% off, what is the n",
          "correct": false,
          "latency_ms": 8711.88473701477
        },
        {
          "q": "Tom is twice as old as Sam. Sam is 5. How old is T",
          "correct": false,
          "latency_ms": 8792.975187301636
        },
        {
          "q": "A rectangle has length 8 and width 5. What is the ",
          "correct": false,
          "latency_ms": 8788.587093353271
        },
        {
          "q": "Sarah reads 15 pages per day. How many pages in 4 ",
          "correct": false,
          "latency_ms": 8676.870346069336
        },
        {
          "q": "If 3x = 12, what is x?",
          "correct": false,
          "latency_ms": 8358.081102371216
        },
        {
          "q": "A pizza is cut into 8 slices. If 3 are eaten, what",
          "correct": false,
          "latency_ms": 8022.942066192627
        }
      ],
      "humaneval": [
        {
          "q": "Write a Python function `add(a, b)` that returns a",
          "correct": false,
          "latency_ms": 8215.983867645264
        },
        {
          "q": "Write a Python function `factorial(n)` that return",
          "correct": false,
          "latency_ms": 7373.898983001709
        },
        {
          "q": "Write a Python function `is_prime(n)` that returns",
          "correct": false,
          "latency_ms": 7252.141952514648
        },
        {
          "q": "Write a Python function `reverse_string(s)` that r",
          "correct": false,
          "latency_ms": 6302.950859069824
        },
        {
          "q": "Write a Python function `fibonacci(n)` that return",
          "correct": false,
          "latency_ms": 6909.268140792847
        }
      ],
      "ifeval": [
        {
          "q": "List exactly 3 fruits, one per line, no extra text",
          "correct": false,
          "latency_ms": 4355.98087310791
        },
        {
          "q": "Respond with ONLY the number 42, nothing else.",
          "correct": true,
          "latency_ms": 5571.604013442993
        },
        {
          "q": "Write a sentence that contains the word 'sovereign",
          "correct": true,
          "latency_ms": 5540.724992752075
        },
        {
          "q": "Output exactly 5 words, all lowercase, describing ",
          "correct": false,
          "latency_ms": 7565.96302986145
        }
      ],
      "latencies": [
        11216.080904006958,
        4791.441917419434,
        7696.846723556519,
        6848.768949508667,
        5726.844787597656,
        8844.170808792114,
        8267.307043075562,
        8229.561805725098,
        9811.07211112976,
        8163.851976394653,
        8787.242889404297,
        8994.663953781128,
        8711.88473701477,
        8792.975187301636,
        8788.587093353271,
        8676.870346069336,
        8358.081102371216,
        8022.942066192627,
        8215.983867645264,
        7373.898983001709,
        7252.141952514648,
        6302.950859069824,
        6909.268140792847,
        4355.98087310791,
        5571.604013442993,
        5540.724992752075,
        7565.96302986145
      ],
      "summary": {
        "mmlu_pro_pct": 50.0,
        "gsm8k_pct": 12.5,
        "humaneval_pct": 0.0,
        "ifeval_pct": 50.0,
        "composite_pct": 29.629629629629626,
        "median_latency_ms": 8163.851976394653,
        "p95_latency_ms": 9811.07211112976
      }
    },
    "qwen3-precise:latest": {
      "mmlu_pro": [
        {
          "q": "Which of the following is a primary function of mi",
          "correct": false,
          "latency_ms": 3289.5431518554688
        },
        {
          "q": "What is the capital of Australia?",
          "correct": true,
          "latency_ms": 1812.8478527069092
        },
        {
          "q": "The derivative of x^2 is:",
          "correct": true,
          "latency_ms": 2077.3561000823975
        },
        {
          "q": "Which element has the atomic number 6?",
          "correct": false,
          "latency_ms": 2139.7922039031982
        },
        {
          "q": "Photosynthesis occurs primarily in:",
          "correct": false,
          "latency_ms": 2161.803960800171
        },
        {
          "q": "The square root of 144 is:",
          "correct": true,
          "latency_ms": 2944.3793296813965
        },
        {
          "q": "Who wrote 'Romeo and Juliet'?",
          "correct": true,
          "latency_ms": 2119.42195892334
        },
        {
          "q": "The chemical symbol for gold is:",
          "correct": true,
          "latency_ms": 1678.4460544586182
        },
        {
          "q": "Speed of light in vacuum (m/s, approx):",
          "correct": true,
          "latency_ms": 1898.1091976165771
        },
        {
          "q": "Which planet is closest to the Sun?",
          "correct": true,
          "latency_ms": 2193.398952484131
        }
      ],
      "gsm8k": [
        {
          "q": "Janet has 3 apples. She buys 5 more. She gives 2 t",
          "correct": true,
          "latency_ms": 2381.458044052124
        },
        {
          "q": "A train travels 60 miles in 1 hour. At the same sp",
          "correct": true,
          "latency_ms": 2850.4209518432617
        },
        {
          "q": "If a shirt costs $20 and is 25% off, what is the n",
          "correct": true,
          "latency_ms": 3592.8540229797363
        },
        {
          "q": "Tom is twice as old as Sam. Sam is 5. How old is T",
          "correct": true,
          "latency_ms": 2461.2340927124023
        },
        {
          "q": "A rectangle has length 8 and width 5. What is the ",
          "correct": true,
          "latency_ms": 3253.8628578186035
        },
        {
          "q": "Sarah reads 15 pages per day. How many pages in 4 ",
          "correct": true,
          "latency_ms": 3505.5270195007324
        },
        {
          "q": "If 3x = 12, what is x?",
          "correct": false,
          "latency_ms": 4124.183893203735
        },
        {
          "q": "A pizza is cut into 8 slices. If 3 are eaten, what",
          "correct": false,
          "latency_ms": 3208.3141803741455
        }
      ],
      "humaneval": [
        {
          "q": "Write a Python function `add(a, b)` that returns a",
          "correct": false,
          "latency_ms": 3545.1698303222656
        },
        {
          "q": "Write a Python function `factorial(n)` that return",
          "correct": false,
          "latency_ms": 4074.3377208709717
        },
        {
          "q": "Write a Python function `is_prime(n)` that returns",
          "correct": false,
          "latency_ms": 3893.812894821167
        },
        {
          "q": "Write a Python function `reverse_string(s)` that r",
          "correct": false,
          "latency_ms": 3804.5361042022705
        },
        {
          "q": "Write a Python function `fibonacci(n)` that return",
          "correct": false,
          "latency_ms": 3880.1002502441406
        }
      ],
      "ifeval": [
        {
          "q": "List exactly 3 fruits, one per line, no extra text",
          "correct": false,
          "latency_ms": 1835.0870609283447
        },
        {
          "q": "Respond with ONLY the number 42, nothing else.",
          "correct": true,
          "latency_ms": 1754.2667388916016
        },
        {
          "q": "Write a sentence that contains the word 'sovereign",
          "correct": true,
          "latency_ms": 3066.8647289276123
        },
        {
          "q": "Output exactly 5 words, all lowercase, describing ",
          "correct": false,
          "latency_ms": 1667.5100326538086
        }
      ],
      "latencies": [
        3289.5431518554688,
        1812.8478527069092,
        2077.3561000823975,
        2139.7922039031982,
        2161.803960800171,
        2944.3793296813965,
        2119.42195892334,
        1678.4460544586182,
        1898.1091976165771,
        2193.398952484131,
        2381.458044052124,
        2850.4209518432617,
        3592.8540229797363,
        2461.2340927124023,
        3253.8628578186035,
        3505.5270195007324,
        4124.183893203735,
        3208.3141803741455,
        3545.1698303222656,
        4074.3377208709717,
        3893.812894821167,
        3804.5361042022705,
        3880.1002502441406,
        1835.0870609283447,
        1754.2667388916016,
        3066.8647289276123,
        1667.5100326538086
      ],
      "summary": {
        "mmlu_pro_pct": 70.0,
        "gsm8k_pct": 75.0,
        "humaneval_pct": 0.0,
        "ifeval_pct": 50.0,
        "composite_pct": 55.55555555555556,
        "median_latency_ms": 2850.4209518432617,
        "p95_latency_ms": 4074.3377208709717
      }
    },
    "qwen3-formal:latest": {
      "mmlu_pro": [
        {
          "q": "Which of the following is a primary function of mi",
          "correct": true,
          "latency_ms": 4723.188877105713
        },
        {
          "q": "What is the capital of Australia?",
          "correct": true,
          "latency_ms": 1152.2901058197021
        },
        {
          "q": "The derivative of x^2 is:",
          "correct": true,
          "latency_ms": 1700.0060081481934
        },
        {
          "q": "Which element has the atomic number 6?",
          "correct": true,
          "latency_ms": 1740.8859729766846
        },
        {
          "q": "Photosynthesis occurs primarily in:",
          "correct": true,
          "latency_ms": 2198.6489295959473
        },
        {
          "q": "The square root of 144 is:",
          "correct": true,
          "latency_ms": 2587.0118141174316
        },
        {
          "q": "Who wrote 'Romeo and Juliet'?",
          "correct": true,
          "latency_ms": 2290.5828952789307
        },
        {
          "q": "The chemical symbol for gold is:",
          "correct": true,
          "latency_ms": 1697.868824005127
        },
        {
          "q": "Speed of light in vacuum (m/s, approx):",
          "correct": true,
          "latency_ms": 3460.839033126831
        },
        {
          "q": "Which planet is closest to the Sun?",
          "correct": true,
          "latency_ms": 3118.1297302246094
        }
      ],
      "gsm8k": [
        {
          "q": "Janet has 3 apples. She buys 5 more. She gives 2 t",
          "correct": true,
          "latency_ms": 3104.3310165405273
        },
        {
          "q": "A train travels 60 miles in 1 hour. At the same sp",
          "correct": true,
          "latency_ms": 3994.594097137451
        },
        {
          "q": "If a shirt costs $20 and is 25% off, what is the n",
          "correct": false,
          "latency_ms": 3623.673915863037
        },
        {
          "q": "Tom is twice as old as Sam. Sam is 5. How old is T",
          "correct": true,
          "latency_ms": 2785.4459285736084
        },
        {
          "q": "A rectangle has length 8 and width 5. What is the ",
          "correct": false,
          "latency_ms": 3523.8349437713623
        },
        {
          "q": "Sarah reads 15 pages per day. How many pages in 4 ",
          "correct": false,
          "latency_ms": 3851.844310760498
        },
        {
          "q": "If 3x = 12, what is x?",
          "correct": false,
          "latency_ms": 3520.72811126709
        },
        {
          "q": "A pizza is cut into 8 slices. If 3 are eaten, what",
          "correct": false,
          "latency_ms": 3331.3381671905518
        }
      ],
      "humaneval": [
        {
          "q": "Write a Python function `add(a, b)` that returns a",
          "correct": false,
          "latency_ms": 3939.465284347534
        },
        {
          "q": "Write a Python function `factorial(n)` that return",
          "correct": false,
          "latency_ms": 3799.633026123047
        },
        {
          "q": "Write a Python function `is_prime(n)` that returns",
          "correct": false,
          "latency_ms": 3910.1619720458984
        },
        {
          "q": "Write a Python function `reverse_string(s)` that r",
          "correct": false,
          "latency_ms": 3991.957902908325
        },
        {
          "q": "Write a Python function `fibonacci(n)` that return",
          "correct": false,
          "latency_ms": 3670.947790145874
        }
      ],
      "ifeval": [
        {
          "q": "List exactly 3 fruits, one per line, no extra text",
          "correct": true,
          "latency_ms": 1780.9269428253174
        },
        {
          "q": "Respond with ONLY the number 42, nothing else.",
          "correct": false,
          "latency_ms": 3785.166025161743
        },
        {
          "q": "Write a sentence that contains the word 'sovereign",
          "correct": false,
          "latency_ms": 2576.1260986328125
        },
        {
          "q": "Output exactly 5 words, all lowercase, describing ",
          "correct": false,
          "latency_ms": 2342.341184616089
        }
      ],
      "latencies": [
        4723.188877105713,
        1152.2901058197021,
        1700.0060081481934,
        1740.8859729766846,
        2198.6489295959473,
        2587.0118141174316,
        2290.5828952789307,
        1697.868824005127,
        3460.839033126831,
        3118.1297302246094,
        3104.3310165405273,
        3994.594097137451,
        3623.673915863037,
        2785.4459285736084,
        3523.8349437713623,
        3851.844310760498,
        3520.72811126709,
        3331.3381671905518,
        3939.465284347534,
        3799.633026123047,
        3910.1619720458984,
        3991.957902908325,
        3670.947790145874,
        1780.9269428253174,
        3785.166025161743,
        2576.1260986328125,
        2342.341184616089
      ],
      "summary": {
        "mmlu_pro_pct": 100.0,
        "gsm8k_pct": 37.5,
        "humaneval_pct": 0.0,
        "ifeval_pct": 25.0,
        "composite_pct": 51.85185185185185,
        "median_latency_ms": 3331.3381671905518,
        "p95_latency_ms": 3994.594097137451
      }
    },
    "qwen25-balanced:latest": {
      "mmlu_pro": [
        {
          "q": "Which of the following is a primary function of mi",
          "correct": true,
          "latency_ms": 2534.7330570220947
        },
        {
          "q": "What is the capital of Australia?",
          "correct": true,
          "latency_ms": 1516.5069103240967
        },
        {
          "q": "The derivative of x^2 is:",
          "correct": false,
          "latency_ms": 3609.015703201294
        },
        {
          "q": "Which element has the atomic number 6?",
          "correct": false,
          "latency_ms": 3240.0691509246826
        },
        {
          "q": "Photosynthesis occurs primarily in:",
          "correct": false,
          "latency_ms": 2151.4439582824707
        },
        {
          "q": "The square root of 144 is:",
          "correct": true,
          "latency_ms": 3470.890998840332
        },
        {
          "q": "Who wrote 'Romeo and Juliet'?",
          "correct": true,
          "latency_ms": 2282.8190326690674
        },
        {
          "q": "The chemical symbol for gold is:",
          "correct": true,
          "latency_ms": 2118.43204498291
        },
        {
          "q": "Speed of light in vacuum (m/s, approx):",
          "correct": true,
          "latency_ms": 2118.1530952453613
        },
        {
          "q": "Which planet is closest to the Sun?",
          "correct": true,
          "latency_ms": 2843.1310653686523
        }
      ],
      "gsm8k": [
        {
          "q": "Janet has 3 apples. She buys 5 more. She gives 2 t",
          "correct": true,
          "latency_ms": 2970.837116241455
        },
        {
          "q": "A train travels 60 miles in 1 hour. At the same sp",
          "correct": false,
          "latency_ms": 3613.831043243408
        },
        {
          "q": "If a shirt costs $20 and is 25% off, what is the n",
          "correct": false,
          "latency_ms": 3658.3712100982666
        },
        {
          "q": "Tom is twice as old as Sam. Sam is 5. How old is T",
          "correct": true,
          "latency_ms": 3435.175895690918
        },
        {
          "q": "A rectangle has length 8 and width 5. What is the ",
          "correct": false,
          "latency_ms": 3794.9609756469727
        },
        {
          "q": "Sarah reads 15 pages per day. How many pages in 4 ",
          "correct": false,
          "latency_ms": 4128.441095352173
        },
        {
          "q": "If 3x = 12, what is x?",
          "correct": false,
          "latency_ms": 3696.4869499206543
        },
        {
          "q": "A pizza is cut into 8 slices. If 3 are eaten, what",
          "correct": false,
          "latency_ms": 3241.057872772217
        }
      ],
      "humaneval": [
        {
          "q": "Write a Python function `add(a, b)` that returns a",
          "correct": false,
          "latency_ms": 3541.5430068969727
        },
        {
          "q": "Write a Python function `factorial(n)` that return",
          "correct": false,
          "latency_ms": 2971.0118770599365
        },
        {
          "q": "Write a Python function `is_prime(n)` that returns",
          "correct": false,
          "latency_ms": 3156.3291549682617
        },
        {
          "q": "Write a Python function `reverse_string(s)` that r",
          "correct": false,
          "latency_ms": 3262.6490592956543
        },
        {
          "q": "Write a Python function `fibonacci(n)` that return",
          "correct": false,
          "latency_ms": 3312.6461505889893
        }
      ],
      "ifeval": [
        {
          "q": "List exactly 3 fruits, one per line, no extra text",
          "correct": true,
          "latency_ms": 1550.473928451538
        },
        {
          "q": "Respond with ONLY the number 42, nothing else.",
          "correct": true,
          "latency_ms": 1625.5757808685303
        },
        {
          "q": "Write a sentence that contains the word 'sovereign",
          "correct": false,
          "latency_ms": 3015.4011249542236
        },
        {
          "q": "Output exactly 5 words, all lowercase, describing ",
          "correct": false,
          "latency_ms": 3170.696973800659
        }
      ],
      "latencies": [
        2534.7330570220947,
        1516.5069103240967,
        3609.015703201294,
        3240.0691509246826,
        2151.4439582824707,
        3470.890998840332,
        2282.8190326690674,
        2118.43204498291,
        2118.1530952453613,
        2843.1310653686523,
        2970.837116241455,
        3613.831043243408,
        3658.3712100982666,
        3435.175895690918,
        3794.9609756469727,
        4128.441095352173,
        3696.4869499206543,
        3241.057872772217,
        3541.5430068969727,
        2971.0118770599365,
        3156.3291549682617,
        3262.6490592956543,
        3312.6461505889893,
        1550.473928451538,
        1625.5757808685303,
        3015.4011249542236,
        3170.696973800659
      ],
      "summary": {
        "mmlu_pro_pct": 70.0,
        "gsm8k_pct": 25.0,
        "humaneval_pct": 0.0,
        "ifeval_pct": 50.0,
        "composite_pct": 40.74074074074074,
        "median_latency_ms": 3170.696973800659,
        "p95_latency_ms": 3794.9609756469727
      }
    },
    "qwen25-creative:latest": {
      "mmlu_pro": [
        {
          "q": "Which of the following is a primary function of mi",
          "correct": false,
          "latency_ms": 3023.613214492798
        },
        {
          "q": "What is the capital of Australia?",
          "correct": true,
          "latency_ms": 2531.738758087158
        },
        {
          "q": "The derivative of x^2 is:",
          "correct": false,
          "latency_ms": 2821.1209774017334
        },
        {
          "q": "Which element has the atomic number 6?",
          "correct": true,
          "latency_ms": 1739.1269207000732
        },
        {
          "q": "Photosynthesis occurs primarily in:",
          "correct": false,
          "latency_ms": 2297.619104385376
        },
        {
          "q": "The square root of 144 is:",
          "correct": false,
          "latency_ms": 3306.3859939575195
        },
        {
          "q": "Who wrote 'Romeo and Juliet'?",
          "correct": true,
          "latency_ms": 1857.5820922851562
        },
        {
          "q": "The chemical symbol for gold is:",
          "correct": true,
          "latency_ms": 2725.8307933807373
        },
        {
          "q": "Speed of light in vacuum (m/s, approx):",
          "correct": false,
          "latency_ms": 2242.778778076172
        },
        {
          "q": "Which planet is closest to the Sun?",
          "correct": true,
          "latency_ms": 1909.9352359771729
        }
      ],
      "gsm8k": [
        {
          "q": "Janet has 3 apples. She buys 5 more. She gives 2 t",
          "correct": true,
          "latency_ms": 1947.7548599243164
        },
        {
          "q": "A train travels 60 miles in 1 hour. At the same sp",
          "correct": false,
          "latency_ms": 3043.4658527374268
        },
        {
          "q": "If a shirt costs $20 and is 25% off, what is the n",
          "correct": true,
          "latency_ms": 2474.1430282592773
        },
        {
          "q": "Tom is twice as old as Sam. Sam is 5. How old is T",
          "correct": true,
          "latency_ms": 2531.237840652466
        },
        {
          "q": "A rectangle has length 8 and width 5. What is the ",
          "correct": true,
          "latency_ms": 2633.7730884552
        },
        {
          "q": "Sarah reads 15 pages per day. How many pages in 4 ",
          "correct": true,
          "latency_ms": 2703.814744949341
        },
        {
          "q": "If 3x = 12, what is x?",
          "correct": false,
          "latency_ms": 2473.097085952759
        },
        {
          "q": "A pizza is cut into 8 slices. If 3 are eaten, what",
          "correct": false,
          "latency_ms": 2132.2031021118164
        }
      ],
      "humaneval": [
        {
          "q": "Write a Python function `add(a, b)` that returns a",
          "correct": false,
          "latency_ms": 2835.6480598449707
        },
        {
          "q": "Write a Python function `factorial(n)` that return",
          "correct": false,
          "latency_ms": 2645.136833190918
        },
        {
          "q": "Write a Python function `is_prime(n)` that returns",
          "correct": false,
          "latency_ms": 2636.02614402771
        },
        {
          "q": "Write a Python function `reverse_string(s)` that r",
          "correct": false,
          "latency_ms": 2490.978002548218
        },
        {
          "q": "Write a Python function `fibonacci(n)` that return",
          "correct": false,
          "latency_ms": 2840.9318923950195
        }
      ],
      "ifeval": [
        {
          "q": "List exactly 3 fruits, one per line, no extra text",
          "correct": true,
          "latency_ms": 1394.970178604126
        },
        {
          "q": "Respond with ONLY the number 42, nothing else.",
          "correct": true,
          "latency_ms": 2346.7860221862793
        },
        {
          "q": "Write a sentence that contains the word 'sovereign",
          "correct": true,
          "latency_ms": 2059.912919998169
        },
        {
          "q": "Output exactly 5 words, all lowercase, describing ",
          "correct": false,
          "latency_ms": 1609.0240478515625
        }
      ],
      "latencies": [
        3023.613214492798,
        2531.738758087158,
        2821.1209774017334,
        1739.1269207000732,
        2297.619104385376,
        3306.3859939575195,
        1857.5820922851562,
        2725.8307933807373,
        2242.778778076172,
        1909.9352359771729,
        1947.7548599243164,
        3043.4658527374268,
        2474.1430282592773,
        2531.237840652466,
        2633.7730884552,
        2703.814744949341,
        2473.097085952759,
        2132.2031021118164,
        2835.6480598449707,
        2645.136833190918,
        2636.02614402771,
        2490.978002548218,
        2840.9318923950195,
        1394.970178604126,
        2346.7860221862793,
        2059.912919998169,
        1609.0240478515625
      ],
      "summary": {
        "mmlu_pro_pct": 50.0,
        "gsm8k_pct": 62.5,
        "humaneval_pct": 0.0,
        "ifeval_pct": 75.0,
        "composite_pct": 48.148148148148145,
        "median_latency_ms": 2490.978002548218,
        "p95_latency_ms": 3043.4658527374268
      }
    }
  }
}