List your company

Mathematical Reasoning on OlympiadBench (pass@1, pass@16)

52.1Pass@1ExTra-N
42.7445.1747.650.03Jun 23, 2026

Evaluation results

MethodLinks
ExTra-NBackbone=Qwen3-1.7B, Variant=Correctness-gated novelty reward only2026.0652.167.1
ExTra (full)Backbone=Qwen3-1.7B2026.0652.167.1
GRPOBackbone=Qwen3-1.7B2026.0649.164.5
ExTra-RBackbone=Qwen3-1.7B, Variant=Entropy-guided prefix regeneration only2026.0648.462.8
DAPOBackbone=Qwen3-1.7B2026.0643.157.9