List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 25 (Throughput/Speedup)

9,043Throughput (tokens/s)Draft-OPD
76.122,404.064,7327,059.94May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Draft-OPDBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=32, Inference Engine=SGLang2026.059,04386.59
DFlashBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=32, Inference Engine=SGLang2026.058,410—6.06
Draft-OPDBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=16, Inference Engine=SGLang2026.057,29776.59
DFlashBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=16, Inference Engine=SGLang2026.056,841—6.06
Draft-OPDBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=32, Inference Engine=SGLang2026.056,645116.42
DFlashBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=32, Inference Engine=SGLang2026.055,985—5.67
Draft-OPDBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=16, Inference Engine=SGLang2026.055,729166.42
Draft-OPDBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=8, Inference Engine=SGLang2026.055,07176.59
DFlashBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=16, Inference Engine=SGLang2026.054,956—5.67
DFlashBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=8, Inference Engine=SGLang2026.054,750—6.06
Draft-OPDBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=32, Inference Engine=SGLang2026.054,718175.32
Draft-OPDBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=8, Inference Engine=SGLang2026.054,127146.42
DFlashBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=32, Inference Engine=SGLang2026.054,014—4.54
DFlashBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=8, Inference Engine=SGLang2026.053,612—5.67
Draft-OPDBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=16, Inference Engine=SGLang2026.053,187165.32
Draft-OPDBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=4, Inference Engine=SGLang2026.052,98486.59
DFlashBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=4, Inference Engine=SGLang2026.052,755—6.06
DFlashBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=16, Inference Engine=SGLang2026.052,738—4.54
Draft-OPDBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=4, Inference Engine=SGLang2026.052,465166.42
DFlashBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=4, Inference Engine=SGLang2026.052,121—5.67
Draft-OPDBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=8, Inference Engine=SGLang2026.052,111135.32
DFlashBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=8, Inference Engine=SGLang2026.051,858—4.54
Draft-OPDBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=4, Inference Engine=SGLang2026.051,229135.32
DFlashBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=4, Inference Engine=SGLang2026.051,086—4.54
Draft-OPDBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=1, Inference Engine=SGLang2026.0596966.59
DFlashBackbone Model=Qwen3-4B (Enable Thinking), Concurrency=1, Inference Engine=SGLang2026.05912—6.06
Draft-OPDBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=1, Inference Engine=SGLang2026.05741126.42
DFlashBackbone Model=Qwen3-8B (Enable Thinking), Concurrency=1, Inference Engine=SGLang2026.05662—5.67
Draft-OPDBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=1, Inference Engine=SGLang2026.05476135.32
DFlashBackbone Model=Qwen3-30B-A3B-Thinking-2507, Concurrency=1, Inference Engine=SGLang2026.05421—4.54