{
  "generated": "2026-09-08",
  "benchmarks": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/benchmarks.json",
  "evaluators": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/evaluators.json",
  "results": {
    "aa-omniscience": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/aa-omniscience.json",
    "agentbench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/agentbench.json",
    "aider-polyglot": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/aider-polyglot.json",
    "aime-2025": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/aime-2025.json",
    "androidworld": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/androidworld.json",
    "arc-agi-1": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/arc-agi-1.json",
    "arc-agi-2": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/arc-agi-2.json",
    "arc-agi-3": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/arc-agi-3.json",
    "automationbench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/automationbench.json",
    "bbh": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/bbh.json",
    "benchcad": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/benchcad.json",
    "bfcl": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/bfcl.json",
    "browsecomp": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/browsecomp.json",
    "cybench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/cybench.json",
    "deepsearchqa": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/deepsearchqa.json",
    "drop": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/drop.json",
    "fieldworkarena": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/fieldworkarena.json",
    "frontiermath": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/frontiermath.json",
    "gaia": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/gaia.json",
    "gdpval": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/gdpval.json",
    "gpqa-diamond": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/gpqa-diamond.json",
    "gsm8k": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/gsm8k.json",
    "healthbench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/healthbench.json",
    "hellaswag": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/hellaswag.json",
    "hle": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/hle.json",
    "humaneval": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/humaneval.json",
    "ifeval": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/ifeval.json",
    "livecodebench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/livecodebench.json",
    "lmarena-text": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/lmarena-text.json",
    "math": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/math.json",
    "mbpp": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/mbpp.json",
    "mle-bench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/mle-bench.json",
    "mmlu": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/mmlu.json",
    "mmlu-pro": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/mmlu-pro.json",
    "mmmu": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/mmmu.json",
    "mmmu-pro": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/mmmu-pro.json",
    "osworld": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/osworld.json",
    "osworld-2": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/osworld-2.json",
    "paperbench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/paperbench.json",
    "ruler": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/ruler.json",
    "screenspot-pro": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/screenspot-pro.json",
    "simpleqa": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/simpleqa.json",
    "swe-bench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/swe-bench.json",
    "swe-bench-multilingual": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/swe-bench-multilingual.json",
    "swe-bench-pro": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/swe-bench-pro.json",
    "swe-bench-verified": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/swe-bench-verified.json",
    "tau-bench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/tau-bench.json",
    "tau2-bench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/tau2-bench.json",
    "terminal-bench": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/terminal-bench.json",
    "tool-decathlon": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/tool-decathlon.json",
    "truthfulqa": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/truthfulqa.json",
    "visualwebarena": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/visualwebarena.json",
    "webarena": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/webarena.json",
    "winogrande": "https://alloevil.github.io/llm-benchmarks-tracker/api/v1/results/winogrande.json"
  },
  "schema": {
    "benchmark.schema.json": "https://alloevil.github.io/llm-benchmarks-tracker/schema/benchmark.schema.json",
    "evaluator.schema.json": "https://alloevil.github.io/llm-benchmarks-tracker/schema/evaluator.schema.json",
    "results.schema.json": "https://alloevil.github.io/llm-benchmarks-tracker/schema/results.schema.json"
  }
}