List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME

90Pass@1NSED (High-Perf open-weight)
-2.24821.70145.6569.599Oct 18, 2024Jan 4, 2025Mar 23, 2025Jun 9, 2025Aug 26, 2025Nov 12, 2025Jan 29, 2026
Updated 6mo ago

Evaluation results

MethodLinks
NSED (High-Perf open-weight)Hardware Class=Enterprise, Est. Cost/Sol=High, Model Scale=70B+2026.0190
DeepSeek-R1 (RL-CoT)Hardware Class=Enterprise, Est. Cost/Sol=Medium2026.0184.2
NSED (Consumer open-weight)Hardware Class=Consumer, Est. Cost/Sol=Low, Model Scale=<20B2026.0184
Gemini-2.5-Pro-06-05Hardware Class=Enterprise, Est. Cost/Sol=High2026.0178.3
Phi-4-mini + Mistral3-3BTraining=CORE2026.0171.42
Phi-4-mini + Mistral3-3B + OracleTraining=SD-E²2026.0154.42
Majority Voting (Qwen-8B)Hardware Class=Consumer, Est. Cost/Sol=Low2026.0154
e3-1.7B + Distill + RLRL Data Size=1216, Evaluation Protocol=8 rollouts avg.2026.0136.67
e3-1.7B + RLRL Data Size=1216, Evaluation Protocol=8 rollouts avg.2026.0136.25
e3-1.7B + InT+ RLRL Data Size=64, Evaluation Protocol=8 rollouts avg.2026.0136.25
Ministral-3-8B-ReasoningTraining=SD-E²2026.0118.5
Qwen2.5-7B-InstructTraining=SD-E²2026.0116.4
PersonaHub-MathSFT Dataset=PersonaHub-Math2024.1013.33
Numina MATHSFT Dataset=Numina MATH2024.1013.33
MATRIX-Gen-ReasonSFT Dataset=MATRIX-Gen-Reason2024.1013.33
FPABackbone=Qwen2.5-Math-7B2025.0912.7
Phi-3-small-8k-InstructTraining=SD-E²2026.0112
Off-RLBackbone=Qwen2.5-Math-7B2025.0911.8
DPOBackbone=Qwen2.5-Math-7B2025.0911.6
KTOBackbone=Qwen2.5-Math-7B2025.0911
A*-POBackbone=Qwen2.5-Math-7B2025.0910.9
BaseBackbone=Qwen2.5-Math-7B2025.0910.8
RPOBackbone=Qwen2.5-Math-7B2025.0910.6
SFTBackbone=Qwen2.5-Math-7B2025.0910.2
FPABackbone=Qwen3-4B-Base2025.0910.1
Tulu 3 CodeSFT Dataset=Tulu 3 Code2024.1010
DPOBackbone=Qwen3-4B-Base2025.099.6
KTOBackbone=Qwen3-4B-Base2025.098.9
RPOBackbone=Qwen3-4B-Base2025.098.5
Off-RLBackbone=Qwen3-4B-Base2025.098.5
A*-POBackbone=Qwen3-4B-Base2025.097.9
SFTBackbone=Qwen3-4B-Base2025.096.7
Camel MATHSFT Dataset=Camel MATH2024.106.67
Magpie ReasonSFT Dataset=Magpie Reason2024.106.67
BaseBackbone=Qwen3-4B-Base2025.096.6
Tulu 3 MATHSFT Dataset=Tulu 3 MATH2024.103.33
BaseBackbone=Llama-3.2-3B-Instruct2025.092.7
Off-RLBackbone=Llama-3.2-3B-Instruct2025.092.3
SFTBackbone=Llama-3.2-3B-Instruct2025.092.2
A*-POBackbone=Llama-3.2-3B-Instruct2025.092.1
FPABackbone=Llama-3.2-3B-Instruct2025.092.1
KTOBackbone=Llama-3.2-3B-Instruct2025.092
DPOBackbone=Llama-3.2-3B-Instruct2025.091.8
RPOBackbone=Llama-3.2-3B-Instruct2025.091.3