List your company

Multi-step Reasoning on ScienceQA

92.72AccuracySceneAlign
74.603279.306684.0188.7134Jan 9, 2026Jan 21, 2026Feb 2, 2026Feb 14, 2026Feb 26, 2026Mar 10, 2026Mar 22, 2026

Evaluation results

MethodLinks
SceneAlignModel=Qwen3-VL-4B2026.0192.72
RFTBackbone=Llama-3.1-8B-Instruct, Setting=zero-shot2026.0392.4
self-rewardingBackbone=Llama-3.1-8B-Instruct, Setting=zero-shot2026.0392.2
CoTBackbone=Llama-3.1-8B-Instruct, Setting=four-shot2026.0391.9
ORACLEBackbone=Llama-3.1-8B-Instruct, Setting=zero-shot2026.0391.7
SFTModel=Qwen3-VL-4B2026.0191.53
BaseModel=Qwen3-VL-4B2026.0191.49
BaseModel=InternVL3-8B2026.0190.77
ToT-SFTBackbone=Llama-3.1-8B-Instruct, Setting=zero-shot2026.0390.7
CoTBackbone=Llama-3.1-8B-Instruct, Setting=zero-shot2026.0390.5
SceneAlignModel=InternVL3-8B2026.0190.15
SFTModel=InternVL3-8B2026.0190.02
SceneAlignModel=Qwen2.5-VL-7B2026.0189.12
SFTModel=Qwen2.5-VL-7B2026.0188.8
BaseModel=Qwen2.5-VL-7B2026.0188.71
ORACLEBackbone=Qwen-2.5-7B-instruct, Setting=zero-shot2026.0387.3
RFTBackbone=Qwen-2.5-7B-instruct, Setting=zero-shot2026.0386.3
SceneAlignModel=Qwen2.5-VL-3B2026.0185.1
CoTBackbone=Qwen-2.5-7B-instruct, Setting=four-shot2026.0384.9
ORACLEBackbone=Mistral-7B-Instruct-v0.3, Setting=zero-shot2026.0383.3
SFTModel=Qwen2.5-VL-3B2026.0183.26
SceneAlignModel=LLaVA-Next-8B2026.0183.25
BaseModel=Qwen2.5-VL-3B2026.0183.19
ToT-SFTBackbone=Qwen-2.5-7B-instruct, Setting=zero-shot2026.0382.9
CoTBackbone=Qwen-2.5-7B-instruct, Setting=zero-shot2026.0382.6
self-rewardingBackbone=Qwen-2.5-7B-instruct, Setting=zero-shot2026.0382.5
SFTModel=LLaVA-Next-8B2026.0181.99
AoTModel=LLaVA-Next-8B2026.0181.25
LLaVA-ReasonerModel=LLaVA-Next-8B2026.0181.1
RFTBackbone=Mistral-7B-Instruct-v0.3, Setting=zero-shot2026.0381
self-rewardingBackbone=Mistral-7B-Instruct-v0.3, Setting=zero-shot2026.0380.2
BaseModel=LLaVA-Next-8B2026.0180.12
ToT-SFTBackbone=Mistral-7B-Instruct-v0.3, Setting=zero-shot2026.0377.4
CoTBackbone=Mistral-7B-Instruct-v0.3, Setting=four-shot2026.0377
CoTBackbone=Mistral-7B-Instruct-v0.3, Setting=zero-shot2026.0375.3