List your company

Scientific Reasoning on GPQA (test) (Multilingual Recovery Metrics)

26.6ARQwen2.5-7B Instruct
11.20815.20419.223.196Jan 14, 2026

Evaluation results

MethodLinks
Qwen2.5-7B InstructSetting=Multi-lingual LLM Baseline2026.0126.620.526.127.526.821.421.428.430.625.5
ACTSetting=With supervised fine-tuning, SFT=true2026.0122.118.120.31922.311.621.419.918.319.2
Task ArithmeticSetting=With supervised fine-tuning, SFT=true2026.0121.215.418.317.420.81013.419.417.917.1
DARE MergingSetting=With supervised fine-tuning, SFT=true2026.0120.81318.119.220.810.713.419.421.617.4
TIES MergingSetting=With supervised fine-tuning, SFT=true2026.0120.111.217.416.117.910.51318.516.515.7
SFTSetting=With supervised fine-tuning, SFT=true2026.0118.312.11921.219.4811.421.919.216.7
Task ArithmeticSetting=Ability Transfer, SFT=false2026.0116.54.21716.313.213.812.513.213.613.4
ACTSetting=Ability Transfer (transfer-only), SFT=false2026.0115.413.613.617.61917.617.4151516
DARE MergingSetting=Ability Transfer, SFT=false2026.0115.27.113.213.813.813.813.41315.213.2
Qwen2.5-7B Math InstructSetting=Multi-lingual LLM Baseline2026.0114.39.61311.611.815.411.812.311.212.3
TIES MergingSetting=Ability Transfer, SFT=false2026.0111.8814.514.112.516.311.215.612.713