CodeSOTAβOriginal site ↗
← ARC-Easy / full sub-1B standard evaluation

Completed / CodeSOTA execution on White

Inspect the measurement.

SlayerLab/Slayer149-ARC · ARC-Easy (full test split, raw accuracy) · 66.6% accuracy

Full benchmark split · 2376 scored examples. Full ARC-Easy test split (2376 examples); FP32, 1024-token input ceiling, raw summed answer-token log likelihood, zero-shot Question/Answer prompt. Character-normalized accuracy reported separately as acc_norm. Native masked-padding adapter uses unpadded row forwards. Distinct from the earlier token-normalized pilot.
Run ID / completed
da7edd3fa6604be87c71349a
17 Oct 2026, 02:57 UTC
Model revision
0e6793ecfb8501dccc1c97dcd944731748eb8d17
Comparison scope
cf50d4b44de811db4aed984c74f66a1eab461af452494ae6c905f9c80d3ffc25
Evaluator SHA-256
ea8141a818424f6e010d35f8e237d2f8682640b6d123ed87438ac01655492c8a
Observed resources
84s elapsed · 0.78 GiB peak torch allocation
Dollar cost: not measured

Download evidence

Inspect individual examples

Read the recorded prediction, reference and outcome for each example.

Scored examples
2,376
Passed
1,582
Failed
794
Metric
accuracy

Examples are downloaded only when requested. Search covers all loaded records; 25 examples are displayed per page. Download all examples · JSONL

Example evidence has not been loaded.