List your company
ResearchFormerly WizWand

Mathematics on AIME25 (Accuracy)

93.33Accuracygpt-oss-puzzle-88B
3.203626.60185073.3982Jan 27, 2026Feb 16, 2026Mar 9, 2026Mar 30, 2026Apr 20, 2026May 11, 2026Jun 1, 2026
Updated 5mo ago

Evaluation results

MethodLinks
gpt-oss-puzzle-88BReasoning effort=High, KV cache precision=FP82026.0293.33
gpt-oss-puzzle-88BReasoning effort=High, KV cache precision=BF162026.0292.92
Regex2026.0491.8
gpt-oss-120BReasoning effort=High, KV cache precision=BF162026.0291.46
BERT-Judgeclean_name=BERT-as-a-Judge2026.0491.4
Lowest CentroidModel=GPT-OSS-120B, Number of Samples=642026.0490
gpt-oss-120BReasoning effort=High, KV cache precision=FP82026.0289.58
BF16Precision=BF16, Backbone=Nemotron 3 Nano2026.0189.1
NVFP4 QADPrecision=NVFP4, Type=QAD, Backbone=Nemotron 3 Nano2026.0187.9
gpt-oss-puzzle-88BReasoning effort=Medium, KV cache precision=BF162026.0286.88
Lowest CentroidModel=Olmo-3.1-Think, Number of Samples=642026.0486.7
gpt-oss-puzzle-88BReasoning effort=Medium, KV cache precision=FP82026.0286.67
NVFP4 PTQPrecision=NVFP4, Type=PTQ, Backbone=Nemotron 3 Nano2026.0185
NVFP4 QATPrecision=NVFP4, Type=QAT, Backbone=Nemotron 3 Nano2026.0183.3
Lowest CentroidModel=QWQ-32B, Number of Samples=642026.0483.3
LLM-Judgebackbone=Qwen-3 0.6B2026.0483
Bottom WindowModel=Olmo-3.1-Think, Number of Samples=642026.0480
DSMoEModel=GPT-OSS-20B2026.0478.6
gpt-oss-120BReasoning effort=Medium, KV cache precision=FP82026.0277.92
gpt-oss-120BReasoning effort=Medium, KV cache precision=BF162026.0276.88
Self-CertaintyModel=Olmo-3.1-Think, Number of Samples=642026.0476.7
Lowest CentroidModel=Qwen3-14B, Number of Samples=642026.0476.7
Tail ConfidenceModel=QWQ-32B, Number of Samples=642026.0476.7
Greedy DecodingModel=GPT-OSS-120B, Number of Samples=642026.0476.7
Pass@1Model=GPT-OSS-120B, Number of Samples=642026.0476.5
Pass@1Model=Olmo-3.1-Think, Number of Samples=642026.0475.7
Tail ConfidenceModel=Olmo-3.1-Think, Number of Samples=642026.0473.3
Self-CertaintyModel=Qwen3-14B, Number of Samples=642026.0473.3
Bottom WindowModel=Qwen3-14B, Number of Samples=642026.0473.3
Bottom WindowModel=GPT-OSS-120B, Number of Samples=642026.0473.3
Pass@1Model=Qwen3-14B, Number of Samples=642026.0471.4
Greedy DecodingModel=Olmo-3.1-Think, Number of Samples=642026.0470
Greedy DecodingModel=Qwen3-14B, Number of Samples=642026.0470
Tail ConfidenceModel=Qwen3-14B, Number of Samples=642026.0470
Self-CertaintyModel=QWQ-32B, Number of Samples=642026.0470
Bottom WindowModel=QWQ-32B, Number of Samples=642026.0470
Self-CertaintyModel=GPT-OSS-120B, Number of Samples=642026.0470
Tail ConfidenceModel=GPT-OSS-120B, Number of Samples=642026.0470
Pass@1Model=QWQ-32B, Number of Samples=642026.0468.5
Orig.Model=GPT-OSS-20B2026.0466.3
gpt-oss-puzzle-88BReasoning effort=Low, KV cache precision=BF162026.0266.25
LobsterBase Model=GPT-OSS:120B, Framework Category=Commercial Framework (multi-agent mode)2026.0463.33
Agent Q-MixBase Model=GPT-OSS:120B2026.0463.33
gpt-oss-puzzle-88BReasoning effort=Low, KV cache precision=FP82026.0262.89
Qwen3-8BInference Mode=think2026.0360
DSMoEModel=Qwen3-30B-Instruct2026.0460
SFTModel=GPT-OSS-20B2026.0460
Greedy DecodingModel=QWQ-32B, Number of Samples=642026.0456.7
SSA-LLM-8BInference Mode=think2026.0356.67
AutoGenBase Model=GPT-OSS:120B, Framework Category=Commercial Framework (multi-agent mode)2026.0453.33
Agent FrameworkBase Model=GPT-OSS:120B, Framework Category=Commercial Framework (multi-agent mode)2026.0453.33
gpt-oss-120BReasoning effort=Low, KV cache precision=FP82026.0251.88
gpt-oss-120BReasoning effort=Low, KV cache precision=BF162026.0250
SFTModel=Qwen3-30B-Instruct2026.0450
Orig.Model=Qwen3-30B-Instruct2026.0446.7
RICEModel=Qwen3-30B-Instruct2026.0446.7
RICEModel=GPT-OSS-20B2026.0446.7
Tail ConfidenceModel=Ministral-3-14B-Instruct, Number of Samples=642026.0446.7
LangGraphBase Model=GPT-OSS:120B, Framework Category=Commercial Framework (multi-agent mode)2026.0446.67
LangGraphBase Model=Gemini-3.1-Flash-Lite, Framework Category=Commercial Framework (multi-agent mode)2026.0446.67
Self-CertaintyModel=Ministral-3-14B-Instruct, Number of Samples=642026.0443.3
G-DesignerBase Model=GPT-OSS:120B, Framework Category=Adaptive topology methods2026.0440
GTDBase Model=Gemini-3.1-Flash-Lite, Framework Category=Adaptive topology methods2026.0440
TopoDIMBase Model=Gemini-3.1-Flash-Lite, Framework Category=Adaptive topology methods2026.0440
AutoGenBase Model=Gemini-3.1-Flash-Lite, Framework Category=Commercial Framework (multi-agent mode)2026.0440
Agent FrameworkBase Model=Gemini-3.1-Flash-Lite, Framework Category=Commercial Framework (multi-agent mode)2026.0440
Agent Q-MixBase Model=Gemini-3.1-Flash-Lite2026.0440
Lowest CentroidModel=Ministral-3-14B-Instruct, Number of Samples=642026.0440
Bottom WindowModel=Ministral-3-14B-Instruct, Number of Samples=642026.0436.7
GPTSwarmBase Model=GPT-OSS:120B, Framework Category=Adaptive topology methods2026.0436.67
MaASBase Model=GPT-OSS:120B, Framework Category=Adaptive topology methods2026.0436.67
TopoDIMBase Model=GPT-OSS:120B, Framework Category=Adaptive topology methods2026.0436.67
GPTSwarmBase Model=Gemini-3.1-Flash-Lite, Framework Category=Adaptive topology methods2026.0436.67
MaASBase Model=Gemini-3.1-Flash-Lite, Framework Category=Adaptive topology methods2026.0436.67
LobsterBase Model=Gemini-3.1-Flash-Lite, Framework Category=Commercial Framework (multi-agent mode)2026.0436.67
AgentDropoutBase Model=GPT-OSS:120B, Framework Category=Adaptive topology methods2026.0433.33
GTDBase Model=GPT-OSS:120B, Framework Category=Adaptive topology methods2026.0433.33
LLM-DebateBase Model=Gemini-3.1-Flash-Lite, Framework Category=Static multi-agent2026.0433.33
G-DesignerBase Model=Gemini-3.1-Flash-Lite, Framework Category=Adaptive topology methods2026.0433.33
LLM-DebateBase Model=GPT-OSS:120B, Framework Category=Static multi-agent2026.0430
Base (direct)Base Model=Gemini-3.1-Flash-Lite, Framework Category=Single-agent baseline2026.0430
Greedy DecodingModel=Ministral-3-14B-Instruct, Number of Samples=642026.0430
Pass@1Model=Ministral-3-14B-Instruct, Number of Samples=642026.0427.5
AgentDropoutBase Model=Gemini-3.1-Flash-Lite, Framework Category=Adaptive topology methods2026.0426.67
LLaDA2.1-minigeneration length=optimal, block length=optimal, denoising steps=optimal2026.0426.67
SSA-LLM-8BInference Mode=no-think2026.0323.33
Base (direct)Base Model=GPT-OSS:120B, Framework Category=Single-agent baseline2026.0423.33
LLaDA2.0-minigeneration length=optimal, block length=optimal, denoising steps=optimal2026.0423.33
TIESGroup=Qwen3-4B-Base, Category=Merged Models2026.0620
DARE + TIESGroup=Qwen3-4B-Base, Category=Merged Models2026.0620
RESMERGEGroup=Qwen3-4B-Base, Category=Merged Models2026.0620
Qwen3-8BInference Mode=no-think2026.0316.67
SDAR-8B-Chatgeneration length=optimal, block length=optimal, denoising steps=optimal2026.0416.67
GRPOGroup=Qwen3-4B-Base, Category=Base and Expert Models2026.0616.67
General-ReasonerGroup=Qwen3-4B-Base, Category=Base and Expert Models2026.0616.67
SpiralGroup=Qwen3-4B-Base, Category=Base and Expert Models2026.0616.67
TSV-MergeGroup=Qwen3-4B-Base, Category=Merged Models2026.0616.67
RAMGroup=Qwen3-4B-Base, Category=Merged Models2026.0616.67
TAGroup=Qwen3-4B-Base, Category=Merged Models2026.0613.33
SDAR-30B-A3Bgeneration length=optimal, block length=optimal, denoising steps=optimal2026.046.67