AI Paper Atlas

Research map / Benchmarks / DPG-Bench

DPG-Bench leaderboard

49 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on DPG-Bench. Best overall: 88.28 (BitDance, NeurIPS 2026).

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.

DPG-Bench · overall

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

#overall ↑MethodPaperTopic
188.28BitDance
Qwen3-14B
BitDance: Scaling Autoregressive Generative Models with Binary Tokens
NeurIPS 2026 · Table 4
Autoregressive and latent generation
287.1Native-UniDDT-XL
Qwen3-1.7B
UniDDT: Unifying Multimodal Understanding and Generation with Decoupled Diffusion Transformer
ECCV 2026 · text
Human and embodied modeling
387.01MoS-L (5B)Mixture of States: Routing Token-Level Dynamics for Multimodal Generation
CVPR 2026 · Table 1
Text-to-image diffusion
486.83UniGen-1.5
Qwen2.5-7B
UniGen-1.5: Enhancing Image Generation and Editing through Reward Unification in RL
CVPR 2026 · Table 2
Text-to-image diffusion
586.69UniReflect
Qwen3-VL-4B
UniReflect: Self-Reflection Tuning for Unified Multimodal Understanding and Generation
ECCV 2026 · Table 4
Multimodal LLMs
686.67VQRAE
Qwen3-0.6B
VQRAE: Representation Quantization Autoencoders for Multimodal Understanding, Generation and Reconstruction
CVPR 2026 · Table 4
Human and embodied modeling
786.38ILLUME-XIlluminating Unified Multimodal Model for Free-form Interleaved Text-Image Generation
ECCV 2026 · Table 5
Text-to-image diffusion
886.27SAR
Infinity-8B
Context-Aware Autoregressive Image Generation for Emerging Reasoning Properties
NeurIPS 2026 · Table 3
Autoregressive and latent generation
986.14Show-o2
Qwen2.5-7B-Instruct
Show-o2: Improved Native Unified Multimodal Models
NeurIPS 2025 · DPG-Bench table
Human and embodied modeling
1085.87ThinkGen (w/ CoT)ThinkGen: Generalized Thinking for Visual Generation
CVPR 2026 · Table 3
Multimodal LLMs
24 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

DPG-Bench · score

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

#score ↑MethodPaperTopic
189Qwen-Image w/ Meta-TTRL
Qwen-Image
Meta-TTRL: A Metacognitive Framework for Self-Improving Test-Time Reinforcement Learning for T2I Generation in Unified Multimodal Models
NeurIPS 2026 · Table 1
Reasoning and chain of thought
287.9PPCL(14B)
Qwen-Image
Pluggable Pruning with Contiguous Layer Distillation for Diffusion Transformers
CVPR 2026 · Table 1
Fast diffusion sampling
387.21Harmon + RecA
Harmon
Reconstruction Alignment Improves Unified Multimodal Models
ICLR 2026 · Table 2
Multimodal LLMs
486.45xLARDSelf-Corrected Image Generation with Explainable Latent Rewards
CVPR 2026 · Table 1
Text-to-image diffusion
585.76ARR-RPO (Gemini 3.1 Pro)
FLUX.1-dev
Auto-Rubric as Reward: From Implicit Preference to Explicit Generative Criteria
NeurIPS 2026 · Table 2
Alignment and preferences
685.28NextStep-1NextStep-1: Toward Autoregressive Image Generation with Continuous Tokens at Scale
ICLR 2026 · Table 2
Text-to-image diffusion
784.76UniFlow (0.6B)UniFlow: A Unified Pixel Flow Tokenizer for Visual Understanding and Generation
ICLR 2026 · Table 3
Multimodal LLMs
883.2+ReflectionLavida-O: Elastic Large Masked Diffusion Models for Unified Multimodal Understanding and Generation
ICLR 2026 · Table 3
Text-to-image diffusion
982.95IOMM-B 512Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training
CVPR 2026 · Table 1
Text-to-image diffusion
1082.4Sparse-LaViDa
LaViDa-O 10.4B
Sparse-LaViDa: Sparse Multimodal Discrete Diffusion Language Models
CVPR 2026 · Table 2
Fast diffusion sampling
5 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

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