List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2025 (Accuracy)

91.67AccuracyHEART
-2.210822.162146.53570.9079Jun 4, 2025Jul 19, 2025Sep 2, 2025Oct 17, 2025Dec 1, 2025Jan 15, 2026Mar 2, 2026
Updated 8mo ago

Evaluation results

MethodLinks
HEARTModel=GPT-5 nano2025.0991.67
HEARTModel=Gemini 2.5 Flash2025.0990.83
Qwen3-235B-A22B-R-TAPArchitecture=MoE, Parameter Count=235B2026.0388.7
HEARTModel=Deepseek-Reasoner2025.0988.33
Gemini2.5-Pro2026.0386.7
Qwen3-235B-A22BArchitecture=MoE, Parameter Count=235B2026.0381.5
AM-Thinking-v1-R-TAPArchitecture=Dense, Parameter Count=32B2026.0379.6
OpenAI-o1 (2024-12-17)2026.0379.2
Qwen3-32B-R-TAPArchitecture=Dense, Parameter Count=32B2026.0378.2
OpenAI-o3-mini (Medium)Configuration=Medium2026.0374.8
AM-Thinking-v1Architecture=Dense, Parameter Count=32B2026.0374.4
Qwen3-32BArchitecture=Dense, Parameter Count=32B2026.0372.9
Nemetron-Ultra-253BArchitecture=Dense, Parameter Count=256B2026.0372.5
DeepSeek-R1Architecture=MoE, Parameter Count=671B2026.0370
ODA-Math-460kBackbone=Qwen3-8B-Base, Training Dataset Size=460k2025.1263.3
ODA-Math-460kBackbone=Qwen2.5-7B-Base, Training Dataset Size=460k2025.1256.7
AM-Thinking (math)Backbone=Qwen3-8B-Base, Training Dataset Size=558k2025.1254.6
VanillaModel=Gemini 2.5 Flash2025.0952.5
HEARTModel=Gemma3 12b Instruct2025.0950.83
AM-Thinking (math)Backbone=Qwen2.5-7B-Base, Training Dataset Size=558k2025.1250
MiroMind-M1-SFTBackbone=Qwen3-8B-Base, Training Dataset Size=719k2025.1247.5
SYNTHETIC-2-SFTBackbone=Qwen3-8B-Base, Training Dataset Size=105k2025.1245.8
OmniThought-0528Backbone=Qwen3-8B-Base, Training Dataset Size=365k2025.1245.4
OpenThoughts3Backbone=Qwen2.5-7B-Base, Training Dataset Size=1.2M2025.1245
VanillaModel=Deepseek-Reasoner2025.0940
OmniThought-0528Backbone=Qwen2.5-7B-Base, Training Dataset Size=365k2025.1240
CyclicReflexBackbone=Qwen3-14B, Decoding Strategy=CyclicReflex2025.0640
DIVERBase Model=DeepSeek-R1-Distill-Qwen-7B2025.0936.9
GRPOBase Model=DeepSeek-R1-Distill-Qwen-7B, Reward Strategy=Clip-higher2025.0935
SYNTHETIC-2-SFTBackbone=Qwen2.5-7B-Base, Training Dataset Size=105k2025.1235
CyclicReflexBackbone=Qwen3-8B, Decoding Strategy=CyclicReflex2025.0633
OriginalBackbone=Qwen3-14B, Decoding Strategy=Original2025.0633
TIPBackbone=Qwen3-14B, Decoding Strategy=TIP2025.0633
DeepMath-103KBackbone=Qwen2.5-7B-Base, Training Dataset Size=309k2025.1231.7
Light-R1-SFTBackbone=Qwen3-8B-Base, Training Dataset Size=79k2025.1231.3
MiroMind-M1-SFTBackbone=Qwen2.5-7B-Base, Training Dataset Size=719k2025.1230
OriginalBackbone=Qwen3-8B, Decoding Strategy=Original2025.0627
Fast-Math-R1-SFTBackbone=Qwen2.5-7B-Base, Training Dataset Size=8k2025.1226.7
Light-R1-SFTBackbone=Qwen2.5-7B-Base, Training Dataset Size=79k2025.1226.7
Fast-Math-R-SFTBackbone=Qwen3-8B-Base, Training Dataset Size=8k2025.1225.8
TIPBackbone=Qwen3-8B, Decoding Strategy=TIP2025.0623
AZR-R-TAPBase Model Type=Base2026.0320.1
AZR-R-TAPBase Model Type=Coder2026.0318.2
MegaScience (math)Backbone=Qwen3-8B-Base, Training Dataset Size=414k2025.1217.9
MegaScience (math)Backbone=Qwen2.5-7B-Base, Training Dataset Size=414k2025.1215
DIVERBase Model=Qwen2.5-7B-Base2025.0912.9
GRPOBase Model=Qwen2.5-7B-Base, Reward Strategy=Clip-higher2025.0912.1
Qwen3-8B-BaseBackbone=Qwen3-8B-Base2025.1210.8
CyclicReflexBackbone=Qwen3-4B, Decoding Strategy=CyclicReflex2025.0610
LIMOBackbone=Qwen3-8B-Base, Training Dataset Size=8172025.128.8
DIVERBase Model=Qwen2.5-Math-1.5B2025.098.3
GRPOBase Model=Qwen2.5-Math-1.5B, Reward Strategy=Clip-higher2025.097.7
TIPBackbone=Qwen3-4B, Decoding Strategy=TIP2025.067
Qwen2.5-7B-BaseBackbone=Qwen2.5-7B-Base2025.126.7
OpenMathInstruct-2Backbone=Qwen2.5-7B-Base, Training Dataset Size=1M2025.125
OriginalBackbone=Qwen3-4B, Decoding Strategy=Original2025.063
LIMOBackbone=Qwen2.5-7B-Base, Training Dataset Size=8172025.121.7
DIVERBase Model=LLaMA-3.1-8B-Instruct2025.091.5
GRPOBase Model=LLaMA-3.1-8B-Instruct, Reward Strategy=Clip-higher2025.091.4