AI Paper Atlas

Research map / Benchmarks / VBench

VBench leaderboard

47 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on VBench. Best Total Score: 86.51 (Pyramid-Flow (FLUX) + TTS, ICCV 2025).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

VBench · Total Score

22 papers, each with its best reported number for its own method; higher is better.

#Total Score ↑MethodPaperTopic
186.51Pyramid-Flow (FLUX) + TTS
Pyramid-Flow (FLUX)
Video-T1: Test-time Scaling for Video Generation
ICCV 2025 · Table 2
Video generation
286.28Wan-1.3B-IPO-3
Wan-1.3B
Dual-IPO: Dual-Iterative Preference Optimization for Text-to-Video Generation
ICLR 2026 · Table 2
Video generation
386.14M4V* (Wan2.1)
Wan2.1
M4V: Multimodal Mamba for Efficient Text-to-Video Generation
CVPR 2026 · Table 1
Video generation
485.46RAVEN + CM-GRPO
Wan2.1-T2V-1.3B
RAVEN: Real-time Autoregressive Video Extrapolation with Consistency-model GRPO
NeurIPS 2026 · Table 1
Video generation
585.45PISCES
HunyuanVideo
PISCES: Annotation-free Text-to-Video Post-Training via Optimal Transport-Aligned Rewards
ICML 2026 · Table 1
Video generation
685DMSampler
Wan 2.1-1.3B
Distillation Models are Good Samplers for Diffusion Reinforcement Learning
ICML 2026 · Table 5
RL from feedback
784.46Bernini
Qwen2.5-VL-7B + Wan2.2-A14B
Bernini: Latent Semantic Planning for Video Diffusion
NeurIPS 2026 · Table 9
Multimodal LLMs
884.31Self Forcing (chunk-wise)
Wan2.1-T2V-1.3B
Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion
NeurIPS 2025 · Comparison with relevant basel
Video generation
984.17Ultra FlashUltra Flash: Scaling Real-Time Streaming Video Generation to High Resolutions
NeurIPS 2026 · Table 1
Video generation
1083.52T2V-Turbo-v2 w/ MG
VideoCrafter2
T2V-Turbo-v2: Enhancing Video Model Post-Training through Data, Reward, and Conditional Guidance Design
ICLR 2025 · Table: Evaluation results on V
Video generation
12 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

VBench · accuracy

16 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
194.76Vision-OPD
Qwen3.5-9B
Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation
NeurIPS 2026 · Table 1
Multimodal LLMs
292.15FOCUS
LLaVA-OneVision-7B
FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering
NeurIPS 2025 · text
Visual question answering
391.1UG-Search
InternVL2.5-8B
Training-free Uncertainty Guidance for Complex Visual Tasks with MLLMs
ECCV 2026 · Table 1
Multimodal LLMs
488.2Mini-o3 (Ours)
Qwen2.5-VL-7B-Instruct
Mini-o3: Scaling Up Reasoning Patterns and Interaction Turns for Visual Search
ICLR 2026 · Table 1
Multimodal LLMs
587.6Gemini-2.5-Flash (3500px) + InSight-o3-vS
Gemini-2.5-Flash
InSight-o3: Empowering Multimodal Foundation Models with Generalized Visual Search
ICLR 2026 · Table 1
Multimodal LLMs
686.4Multi-turn ViGoRL-7B
Qwen2.5-VL-7B
Grounded Reinforcement Learning for Visual Reasoning
NeurIPS 2025 · Table 2
Multimodal LLMs
785.9DCRL
Qwen3-VL-8B
Eliciting Complex Spatial Reasoning in MLLMs through Wide-Baseline Matching
CVPR 2026 · Table 5
Multimodal LLMs
884.8CodeDance-7B
Qwen2.5-VL-7B
CodeDance: A Dynamic Tool-integrated MLLM for Executable Visual Reasoning
CVPR 2026 · Table 1
Multimodal LLMs
984.3Pixel Reasoner
Qwen2.5-VL-7B
Pixel Reasoner: Incentivizing Pixel Space Reasoning via Curiosity-Driven Reinforcement Learning
NeurIPS 2025 · Main Results table
Multimodal LLMs
1083.25Ours (Multistage SFT + DPO)
Qwen2.5-VL-7B-Instruct
Long Grounded Thoughts: Synthesizing Visual Problems and Reasoning Chains at Scale
ICML 2026 · Table 2
Multimodal LLMs
6 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

VBench · Semantic Score

13 papers, each with its best reported number for its own method; higher is better.

#Semantic Score ↑MethodPaperTopic
189.11Wan2.1-1.3B + FlowMo
Wan2.1-1.3B
FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation
NeurIPS 2025 · Table 1
Video generation
284.83CogVideoX-5B + TTS
CogVideoX-5B
Video-T1: Test-time Scaling for Video Generation
ICCV 2025 · Table 2
Video generation
381.28Self Forcing (chunk-wise)
Wan2.1-T2V-1.3B
Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion
NeurIPS 2025 · Comparison with relevant basel
Video generation
480.7PyramidalWan (20-20-10)
Wan2.1-1.3B
PyramidalWan: On Making Pretrained Video Model Pyramidal for Efficient Inference
CVPR 2026 · Table 3
Video generation
580.45M4V* (Wan2.1)
Wan2.1
M4V: Multimodal Mamba for Efficient Text-to-Video Generation
CVPR 2026 · Table 1
Video generation
680.33PISCES
HunyuanVideo
PISCES: Annotation-free Text-to-Video Post-Training via Optimal Transport-Aligned Rewards
ICML 2026 · Table 1
Video generation
778.5Presto
Allegro
Long Video Diffusion Generation with Segmented Cross-Attention and Content-Rich Video Data Curation
CVPR 2025 · Table 1
Video generation
878.33T2V-Turbo-v2 w/o MG
VideoCrafter2
T2V-Turbo-v2: Enhancing Video Model Post-Training through Data, Reward, and Conditional Guidance Design
ICLR 2025 · Table: Evaluation results on V
Video generation
974.76T2V-Turbo (VC2), 4 steps
VideoCrafter2
T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback
NeurIPS 2024 · Table 1
Video generation
1073.15APT, 1 stepDiffusion Adversarial Post-Training for One-Step Video Generation
ICML 2025 · Table: Quantitative metrics on
Fast diffusion sampling
3 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

VBench · Quality Score

10 papers, each with its best reported number for its own method; higher is better.

#Quality Score ↑MethodPaperTopic
187.56M4V* (Wan2.1)
Wan2.1
M4V: Multimodal Mamba for Efficient Text-to-Video Generation
CVPR 2026 · Table 1
Video generation
287.5Pyramid-Flow (FLUX) + TTS
Pyramid-Flow (FLUX)
Video-T1: Test-time Scaling for Video Generation
ICCV 2025 · Table 2
Video generation
386.73PISCES
HunyuanVideo
PISCES: Annotation-free Text-to-Video Post-Training via Optimal Transport-Aligned Rewards
ICML 2026 · Table 1
Video generation
485.25Self Forcing (frame-wise)
Wan2.1-T2V-1.3B
Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion
NeurIPS 2025 · Comparison with relevant basel
Video generation
585.13T2V-Turbo-v2 w/ MG
VideoCrafter2
T2V-Turbo-v2: Enhancing Video Model Post-Training through Data, Reward, and Conditional Guidance Design
ICLR 2025 · Table: Evaluation results on V
Video generation
684.74Pyramidal Flow Matching
SD3 Medium
Pyramidal Flow Matching for Efficient Video Generative Modeling
ICLR 2025 · Table 1
Video generation
784.21APT, 1 stepDiffusion Adversarial Post-Training for One-Step Video Generation
ICML 2025 · Table: Quantitative metrics on
Fast diffusion sampling
882.56ANSE (+Ours)
CogVideoX-2B
Model Already Knows the Best Noise: Bayesian Active Noise Selection via Attention in Video Diffusion Model
ICLR 2026 · Table 1
Video generation
980.05FreqPrior
AnimateDiff
FreqPrior: Improving Video Diffusion Models with Frequency Filtering Gaussian Noise
ICLR 2025 · Main results table
Video generation
1073.58Wan2.1-1.3B + FlowMo
Wan2.1-1.3B
FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation
NeurIPS 2025 · Table 1
Video generation

VBench · Dynamic Degree

10 papers, each with its best reported number for its own method; higher is better.

#Dynamic Degree ↑MethodPaperTopic
1100Presto
Allegro
Long Video Diffusion Generation with Segmented Cross-Attention and Content-Rich Video Data Curation
CVPR 2025 · Table 1
Video generation
296.7M4V* (Wan2.1)
Wan2.1
M4V: Multimodal Mamba for Efficient Text-to-Video Generation
CVPR 2026 · Table 1
Video generation
381.96Wan2.1-1.3B + FlowMo
Wan2.1-1.3B
FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation
NeurIPS 2025 · Table 1
Video generation
466.39CogVideoX-2BCogVideoX: Text-to-Video Diffusion Models with An Expert Transformer
ICLR 2025 · Evaluation results table
Video generation
564.63Pyramidal Flow Matching
SD3 Medium
Pyramidal Flow Matching for Efficient Video Generative Modeling
ICLR 2025 · Table 1
Video generation
652.8ARLON
OpenSora-V1.2
ARLON: Boosting Diffusion Transformers with Autoregressive Models for Long Video Generation
ICLR 2025 · Table 2
Video generation
744.44Show-1Show-1: Marrying Pixel and Latent Diffusion Models for Text-to-Video Generation
IJCV 2024 · Table 4
Video generation
843.5VideoCrafter-2.0 fine-tuned on VMChill-2M
VideoCrafter-2.0
VMChill: A Dataset for Fine-Grained Visual-Musical Synergy
AAAI 2026 · Table 3
Video-language understanding
928.06LVD-2M fine-tuning
1.75B T2V diffusion model
LVD-2M: A Long-take Video Dataset with Temporally Dense Captions
NeurIPS 2024 · VBench evaluation table
Video generation
1023.27NOVA
NOVA 0.6B
Autoregressive Video Generation without Vector Quantization
ICLR 2025 · Appendix, Table: Performance c
Video generation

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.