List your company
ResearchFormerly WizWand

Multi-modal Reasoning on EMMA

38.5AccuracyQwen2.5-VL-Instruct
7.09215.24623.431.554Oct 2, 2025Nov 13, 2025Dec 26, 2025Feb 6, 2026Mar 21, 2026May 2, 2026Jun 14, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Qwen2.5-VL-InstructSize=72B2026.0538.5——
Claude-3.5Size=-2026.0538.1——
AnE-3rdTraining Stage=Round 32026.0534.6——
AnE-2ndTraining Stage=Round 22026.0533——
GPT-4oModel Category=Closed-Source MLLMs2026.0132.7——
GPT-4oSize=-2026.0532.7——
PTA-GRPOBase Model=Qwen2.5-7B-VL2025.1031.9——
STRIDEBase model=Qwen2.5-VL-7B2026.0631.5——
MoCASize=7B2026.0531.3——
ReLaX-VL-7BModel Category=Reasoning Multimodal LLM, Parameters=7B2025.1230.6——
AnE-1stTraining Stage=Round 12026.0530.5——
Preliminary RLTraining Phase=Warm-up2026.0530.1——
VL-Rethinker-7BModel Category=Reasoning Multimodal LLM, Parameters=7B2025.1229.7——
VL-RethinkerSize=7B2026.0529.7——
VL-RethinkerEvaluation Source=original paper2026.0529.7——
SRPO-7BModel Category=Reasoning Multimodal LLM, Parameters=7B2025.1229.6——
SRPOEvaluation Source=original paper2026.0529.6——
SRPOBase Model=Qwen2.5-7B-VL2025.1029.6——
SRPOBase model=Qwen2.5-VL-7B2026.0629.6——
DAPO (Ours)Model Category=Multi-image/Video Enhancing MLLMs, Training Strategy=DAPO2026.0129.3——
Two-stage RL (Ours)Model Category=Multi-image/Video Enhancing MLLMs, Training Strategy=Two-stage RL (DPS and annealing)2026.0128.6——
Metis-RISEEvaluation Source=reproduced2026.0528.5——
SPARK-VL-7BEvaluation Source=reproduced2026.0528.5——
DPS (Ours)Model Category=Multi-image/Video Enhancing MLLMs, Training Strategy=DPS2026.0128.4——
LLaVA-Critic-R1Evaluation Source=original paper2026.0528.3——
GPT-4o-miniSize=-2026.0527.3——
ReLaX-VL-3BModel Category=Reasoning Multimodal LLM, Parameters=3B2025.1226.9——
OpenVLThinkerEvaluation Source=original paper2026.0526.8——
OpenVLThinker-7BModel Category=Reasoning Multimodal LLM, Parameters=7B2025.1226.6——
mPLUG-Owl3Model Category=Multi-image/Video Enhancing MLLMs, Parameter Scale=8B2026.0124.8——
Qwen2.5-VL-7B-InstructEvaluation Source=original paper2026.0524.6——
OpenMMReasoner-7BEvaluation Source=reproduced2026.0524.5——
Vision-R1Evaluation Source=reproduced2026.0523.6——
MM-Eureka-7BModel Category=Reasoning Multimodal LLM, Parameters=7B2025.1223.5——
MM-EurekaBase model=Qwen2.5-VL-7B2026.0623.5——
TW-GRPOModel Category=Multi-image/Video Enhancing MLLMs, Parameter Scale=7B2026.0122.5——
Vision-R1-7BModel Category=Reasoning Multimodal LLM, Parameters=7B2025.1222.4——
Qwen2.5-VL-7BModel Category=General Multimodal LLM, Parameters=7B2025.1221.5——
MM-Eureka-8BModel Category=Reasoning Multimodal LLM, Parameters=8B2025.1221.5——
Qwen2.5-VL-InstructSize=7B2026.0521.5——
BaseBase Model=Qwen2.5-7B-VL2025.1021.5——
InternVL2.5Model Category=Open-Source General MLLMs, Parameter Scale=8B2026.0121——
Intern2.5-VL-8BModel Category=General Multimodal LLM, Parameters=8B2025.1220.6——
Qwen2.5-VLModel Category=Open-Source General MLLMs, Parameter Scale=7B2026.0120.4——
Mantis-Idefics2Model Category=Multi-image/Video Enhancing MLLMs, Parameter Scale=8B2026.0120.3——
Qwen2-VL-7BModel Category=General Multimodal LLM, Parameters=7B2025.1220.2——
BaseBase model=Qwen2.5-VL-7B2026.0620.2——
Intern2-VL-8BModel Category=General Multimodal LLM, Parameters=8B2025.1219.8——
Pixel ReasonerSize=7B2026.0519.8——
Revisual-R1Evaluation Source=reproduced2026.0519.5——
Qwen2.5-VL-3BModel Category=General Multimodal LLM, Parameters=3B, reproduced=true2025.1219.2——
LLaVA-NeXT-InterleaveModel Category=Multi-image/Video Enhancing MLLMs, Parameter Scale=7B2026.0119——
MMR1Evaluation Source=reproduced2026.0518.6——
Llava-OV-7BModel Category=General Multimodal LLM, Parameters=7B2025.1218.3——
Llava-OVSize=7B2026.0518.3——
DeepEyesSize=7B2026.0518.1——
VideoRFTModel Category=Multi-image/Video Enhancing MLLMs, Parameter Scale=7B2026.0117.8——
mPLUG-Owl3Size=7B2026.0515——
DocopilotSize=8B2026.0512.1——
R1-VL-7BModel Category=Reasoning Multimodal LLM, Parameters=7B2025.128.3——
R1-VLSize=7B2026.058.3——
BaseBase Model=Qwen3-VL-8B-Instruct2026.05—38.34—
BaseBase Model=Qwen3-VL-2B-Instruct2026.05—26.87—
BaseBase Model=Qwen2.5-VL-7B-Instruct2026.05—23.25—
BaseBase Model=Qwen2-VL-7B-Instruct2026.05—11.75—
Base (Qwen2-VL-8B Instruct)Temperature=0.62026.05—38.34—
Claude-3.5Size=-2026.05——38.1
DAPOBase Model=Qwen2-VL-8B Instruct, Temperature=0.62026.05—38.75—
DeepEyesSize=7B2026.05——18.1
DocopilotSize=8B2026.05——12.1
GPT-4oSize=-2026.05——32.7
GPT-4o-miniSize=-2026.05——27.3
GRPOBase Model=Qwen3-VL-8B-Instruct2026.05—41.34—
GRPOBase Model=Qwen3-VL-2B-Instruct2026.05—28.5—
Llava-OVSize=7B2026.05——18.3
MiMO-VL-InstructSize=7B2026.05——26.5
MINT-CoTBase Model=Qwen2-VL-7B-Instruct2026.05—19—
MM-EurekaBase Model=Qwen2.5-VL-7B-Instruct2026.05—28.75—
mPLUG-Owl3Size=7B2026.05——15
OpenVLThinkerSize=7B2026.05——26.6
PAPOBase Model=Qwen3-VL-8B-Instruct2026.05—39—
PAPOBase Model=Qwen3-VL-2B-Instruct2026.05—27.25—
PAPO_DBase Model=Qwen2-VL-8B Instruct, Temperature=0.62026.05—42.25—
Pixel ReasonerSize=7B2026.05——19.8
Qwen2.5-VL-InstructSize=72B2026.05——38.5
Qwen2.5-VL-InstructSize=7B2026.05——21.5
R1-ShareVLBase Model=Qwen2.5-VL-7B-Instruct2026.05—30.5—
R1-VLSize=7B2026.05——8.3
RAPOBase Model=Qwen3-VL-8B-Instruct2026.05—42.5—
RAPOBase Model=Qwen3-VL-2B-Instruct2026.05—31.25—
RAPO_DBase Model=Qwen2.5-VL-7B-Instruct2026.05—30.5—
RAPO_DBase Model=Qwen2-VL-7B-Instruct2026.05—28.75—
RAPO_DBase Model=Qwen2-VL-8B Instruct, Temperature=0.62026.05—45.5—
RAPO_GBase Model=Qwen2.5-VL-7B-Instruct2026.05—29—
RAPO_GBase Model=Qwen2-VL-7B-Instruct2026.05—25.25—
TVCBase Model=Qwen2-VL-7B-Instruct2026.05—20.75—
VL-RethinkerBase Model=Qwen2.5-VL-7B-Instruct2026.05—28.75—
VL-RethinkerSize=7B2026.05——29.7
VL-ScalerSize=7B2026.05——31.3
VL-Scaler-MiMOSize=7B2026.05——32.1