List your company
ResearchFormerly WizWand

First Integral Prediction on First Integral Discovery Hard (test)

63.7AccuracyFISolver-Base
-0.05216.49933.0549.601May 20, 2026
Updated 5mo ago

Evaluation results

MethodLinks
FISolver-Basebeam size (bs)=70, Training Strategy=RL, RL Training Data=HardSplit (HS), RL Reward Scheme=LD-based Reward (LD)2026.0563.7
FISolver-Basebeam size (bs)=70, Training Strategy=RL, RL Training Data=HardSplit (HS), RL Reward Scheme=Binary Reward (Bin)2026.0563.3
FISolver-Basebeam size (bs)=70, Training Strategy=SFT-only2026.0562.2
FISolver-Basebeam size (bs)=70, Training Strategy=RL, RL Training Data=NormalSplit (NS), RL Reward Scheme=LD-based Reward (LD)2026.0561.5
FISolver-Basebeam size (bs)=70, Training Strategy=RL, RL Training Data=NormalSplit (NS), RL Reward Scheme=Binary Reward (Bin)2026.0559.9
FISolver-Basebeam size (bs)=5, Training Strategy=RL, RL Training Data=HardSplit (HS), RL Reward Scheme=LD-based Reward (LD)2026.0533.5
FISolver-Basebeam size (bs)=5, Training Strategy=RL, RL Training Data=NormalSplit (NS), RL Reward Scheme=LD-based Reward (LD)2026.0531
FISolver-Basebeam size (bs)=5, Training Strategy=RL, RL Training Data=NormalSplit (NS), RL Reward Scheme=Binary Reward (Bin)2026.0529.3
FISolver-Basebeam size (bs)=5, Training Strategy=SFT-only2026.0526.1
FISolver-Basebeam size (bs)=5, Training Strategy=RL, RL Training Data=HardSplit (HS), RL Reward Scheme=Binary Reward (Bin)2026.0526
Mathematica2026.0523.3
Qwen2.5-Math-7Bbeam size (bs)=52026.052.6
Qwen2.5-Math-1.5Bbeam size (bs)=52026.052.4