ResearchBenchmarksMath Reasoning on GSM8K (BA@20/40/60)Follow95.2BA Accuracy @20RoRo92.28893.04493.894.556May 28, 2026Evaluation ResultsMethodMethodLinksBA Accuracy @20BA Accuracy @40BA Accuracy @60RoRoSetting=Same-family, M...Setting=Same-family, Models=Qwen3-1.7B / Qwen3-14B2026.0595.296.196.5RoRoSetting=Cross-family,...Setting=Cross-family, Models=Qwen3-1.7B / DeepSeek-R1-Distill-Qwen-14B2026.0592.493.694.2