List your company

Scientific and Tool-Use Reasoning on SciKnowEval and ToolUse In-domain

81.45Material ScoreGRPO
58.840464.710270.5876.4498May 27, 2026

Evaluation results

MethodLinks
GRPOBackbone=Qwen3-4B2026.0581.4574.6153.7577.5361.9569.86
ROSDBackbone=Qwen3-4B2026.0580.1876.5657.582.4767.4672.83
ROSDBackbone=Qwen3-8B2026.0579.5275.1660.3883.0769.1273.45
SDPOBackbone=Qwen3-8B2026.0578.4676.4161.1380.7165.8172.5
GRPOBackbone=Qwen3-8B2026.0577.4676.0956.8880.7768.7571.99
SDPOBackbone=Qwen3-4B2026.0573.668.5255.3876.9360.6667.02
Base ModelBackbone=Qwen3-4B2026.0562.960.6331.6342.9258.5551.32
Base ModelBackbone=Qwen3-8B2026.0559.7159.0632.7541.2857.7250.1