II.
TestSet JSON
Structured · livetest-set:gaia-validation
GAIA validation split json
Inspect the normalized record payload exactly as the atlas UI reads it.
{
"id": "test-set:gaia-validation",
"_kind": "TestSet",
"_file": "benchmarks/test-sets/gaia-validation.yaml",
"_cluster": "benchmarks",
"attributes": {
"displayName": "GAIA validation split",
"benchmarkId": "benchmark:gaia",
"splitName": "validation",
"itemCount": 165,
"description": "Validation split of the GAIA benchmark (Mialon et al., 2023). 165\nheld-out questions across three difficulty levels. Used as the\npublic-leaderboard split because the test split is hidden.\n",
"sourceUrl": "https://huggingface.co/datasets/gaia-benchmark/GAIA"
},
"outgoingEdges": [
{
"from": "test-set:gaia-validation",
"to": "benchmark:gaia",
"kind": "split_of",
"attributes": {}
}
],
"incomingEdges": [
{
"from": "eval-run:gaia.claude-code.2025",
"to": "test-set:gaia-validation",
"kind": "uses_test_set"
}
]
}