ResearchBenchmarksNovelty Alignment unseen (test)Follow78Accuracyo3-mini31.243.3555.567.65Jan 16, 2026Evaluation ResultsMethodMethodLinksAccuracyo3-mini2026.0178SciRM-Ref2026.0174Qwen2.52026.0171SciRM2026.0161Qwen32026.0158Selene2026.0157Llama2026.0156Granite2026.0148Skywork2026.0142DS-Qwen2026.0141DS-GRM2026.0135Prometheus2026.0133