Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Arithmetic Reasoning on Matchstick Arithmetic (held-out)
Loading...
90.7
Accuracy
CORE
57.732
66.291
74.85
83.409
May 27, 2026
Accuracy
Updated 6d ago
Evaluation Results
Method
Method
Links
Accuracy
CORE
training samples=10
2026.05
90.7
CORE
training samples=5
2026.05
87.3
CORE
training samples=100
2026.05
87
GEPA
training samples=100
2026.05
77
Episodic RAG
training samples=5
2026.05
70.3
MemRL
training samples=100
2026.05
70.3
MemRL
training samples=5
2026.05
70
GEPA
training samples=10
2026.05
69.3
GEPA
training samples=5
2026.05
68.7
No Learning
training samples=5
2026.05
68.1
No Learning
training samples=10
2026.05
68.1
No Learning
training samples=100
2026.05
68.1
MemRL
training samples=10
2026.05
64.7
Episodic RAG
training samples=100
2026.05
64
GRPO
training samples=10
2026.05
63.7
GRPO
training samples=5
2026.05
63
Episodic RAG
training samples=10
2026.05
62.7
GRPO
training samples=100
2026.05
59
Feedback
Search any
task
Search any
task