List your company

Arithmetic Reasoning on Matchstick Arithmetic (held-out)

90.7AccuracyCORE
57.73266.29174.8583.409May 27, 2026

Evaluation results

MethodLinks
COREtraining samples=102026.0590.7
COREtraining samples=52026.0587.3
COREtraining samples=1002026.0587
GEPAtraining samples=1002026.0577
Episodic RAGtraining samples=52026.0570.3
MemRLtraining samples=1002026.0570.3
MemRLtraining samples=52026.0570
GEPAtraining samples=102026.0569.3
GEPAtraining samples=52026.0568.7
No Learningtraining samples=52026.0568.1
No Learningtraining samples=102026.0568.1
No Learningtraining samples=1002026.0568.1
MemRLtraining samples=102026.0564.7
Episodic RAGtraining samples=1002026.0564
GRPOtraining samples=102026.0563.7
GRPOtraining samples=52026.0563
Episodic RAGtraining samples=102026.0562.7
GRPOtraining samples=1002026.0559