List your company
ResearchFormerly WizWand

Mathematical Reasoning on WeMath 525 samples

78.5AccuracyDoubao-Seed-1.6
33.15644.92856.768.472May 8, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Doubao-Seed-1.6Model Category=Closed-Source SOTA Models2026.0578.5
Qwen3-VL-235B-ThinkingModel Category=Open-Source Large Baselines2026.0574.9
Gemini3-proModel Category=Closed-Source SOTA Models2026.0573.8
GPT-5-20250807Model Category=Closed-Source SOTA Models2026.0571
GLM-4.5VModel Category=Closed-Source SOTA Models2026.0568.8
OSTBase Model=Qwen3-VL-8B-Instruct, Sampling Ratio=Best-20%2026.0559
DEITABase Model=Qwen3-VL-8B-Instruct, Sampling Ratio=Top-20%2026.0558.6
Qwen3-VL-8B-Instruct + LLM-as-a-JudgeSampling Ratio=20%2026.0558.1
Qwen3-VL-8B-InstructStrategy=Base2026.0557.5
Qwen3-VL-8B-Instruct + RandomSampling Ratio=20%2026.0556.8
OSTBase Model=Qwen3-VL-4B-Instruct, Sampling Ratio=Best-20%2026.0556
DEITABase Model=Qwen3-VL-4B-Instruct, Sampling Ratio=Top-20%2026.0555.2
Qwen3-VL-8B-Instruct + Full SFTSampling Ratio=100%2026.0555.2
Qwen3-VL-4B-Instruct + LLM-as-a-JudgeSampling Ratio=20%2026.0555
Qwen3-VL-4B-InstructStrategy=Base2026.0554.3
Qwen3-VL-4B-Instruct + RandomSampling Ratio=20%2026.0553.9
Qwen3-VL-4B-Instruct + Full SFTSampling Ratio=100%2026.0552.6
Kimi-vl-A3B-thinkingModel Category=Open-Source Large Baselines2026.0547
OSTBase Model=Qwen3-VL-2B-Instruct, Sampling Ratio=Best-20%2026.0537.5
DEITABase Model=Qwen3-VL-2B-Instruct, Sampling Ratio=Top-20%2026.0536.5
Qwen3-VL-2B-Instruct + LLM-as-a-JudgeSampling Ratio=20%2026.0536.2
Qwen3-VL-2B-InstructStrategy=Base2026.0535.8
Qwen3-VL-2B-Instruct + RandomSampling Ratio=20%2026.0535.4
Qwen3-VL-2B-Instruct + Full SFTSampling Ratio=100%2026.0534.9