AI Paper Atlas

Research map / Benchmarks / VQAv2

VQAv2 leaderboard

213 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on VQAv2. Best accuracy: 87.7 (AIM, 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.

VQAv2 · accuracy

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

102 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

VQAv2 · score

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

#score ↑MethodPaperTopic
183.6PaliGemma-2B (ViLex@224)
PaliGemma-2B
Visual Lexicon: Rich Image Features in Language Space
CVPR 2025 · Table 5
CLIP and zero-shot
282.9HAMSTER
VILA1.5-13B
HAMSTER: Hierarchical Action Models for Open-World Robot Manipulation
ICLR 2025 · Table: Core VQA and robustness
Robot action and navigation
382.5VLLMv2+IOP
VLLMv2
IOP: An Idempotent-Like Optimization Method on the Pareto Front of Hypernetwork
AAAI 2025 · Table 5
Recognition and long-tail
482.1UIX-Qwen2
Qwen2-7B-Instruct
Harnessing Webpage UIs for Text-Rich Visual Understanding
ICLR 2025 · Table: Performance on other ge
Multimodal LLMs
581.8InternVL2-8B + RP
InternVL2-8B
Img-Diff: Contrastive Data Synthesis for Multimodal Large Language Models
CVPR 2025 · Table 3
Multimodal LLMs
681.6LLaVA-S² (Ours)
LLaVA-S², Vicuna-7B
DenseFusion-1M: Merging Vision Experts for Comprehensive Multimodal Perception
NeurIPS 2024 · Main results table
Multimodal LLMs
781.3Mipha-3B
Phi-2 (2.7B)
A Comprehensive Overhaul of Multimodal Assistant with Small Language Models
AAAI 2025 · Table: Multi-modal evaluation
Multimodal LLMs
881.1HaploVL-7B-Pro
Qwen2.5-7B
HaploVL: A Single-Transformer Baseline for Multi-Modal Understanding
ICML 2025 · Main results table
Language, safety and interpretability
981.1SpatialRGPT-VILA-1.5-3B
VILA-1.5-3B
SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models
NeurIPS 2024 · Table 2
Multimodal LLMs
1080.9AIMV2-3B
ViT-3B/14
Multimodal Autoregressive Pre-training of Large Vision Encoders
CVPR 2025 · Table 7
CLIP and zero-shot
28 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

VQAv2 · accuracy (test-dev)

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

#accuracy ↑MethodPaperTopic
185.7Omni-SMoLA PaLI-X FT, per-task LoRA
Omni-SMoLA PaLI-X FT
Omni-SMoLA: Boosting Generalist Multimodal Models with Soft Mixture of Low-rank Experts
CVPR 2024 · Table 7
Multimodal LLMs
284.3PaLI-17B
mT5-XXL + ViT-e
PaLI: A Jointly-Scaled Multilingual Language-Image Model
ICLR 2023 · Table 2
Human and embodied modeling
381.7Florence-2-LFlorence-2: Advancing a Unified Representation for a Variety of Vision Tasks
CVPR 2024 · Table 4
CLIP and zero-shot
480.8VisionLLM v2
Vicuna-7B
VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks
NeurIPS 2024 · Table 1
Multimodal LLMs
580.3LOVA³-7B
LLaVA-1.5 / Vicuna-7B
LOVA3: Learning to Visual Question Answering, Asking and Assessment
NeurIPS 2024 · Table: Results on five generic
Visual question answering
680PaLM-E-562B
PaLM-E-562B
PaLM-E: An Embodied Multimodal Language Model
ICML 2023 · Table 4
Robot action and navigation
779.1VistaLLM-13B
Vicuna-13B
Jack of All Tasks Master of Many: Designing General-Purpose Coarse-to-Fine Vision-Language Model
CVPR 2024 · Table 2
Multimodal LLMs
877.9Unified-IO XLUNIFIED-IO: A Unified Model for Vision, Language, and Multi-modal Tasks
ICLR 2023 · Table 4
Human and embodied modeling
977.69Aurora (ours, r=64)
BLIP-base
Parameter-efficient Tuning of Large-scale Multimodal Foundation Model
NeurIPS 2023 · Table 3
CLIP and zero-shot
1077.68METER-CLIP-ViT_BASE
CLIP-ViT_BASE + RoBERTa-base
An Empirical Study of Training End-to-End Vision-and-Language Transformers
CVPR 2022 · Table 7
Human and embodied modeling
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

VQAv2 · accuracy (val)

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

9 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

VQAv2 · accuracy (test-std)

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

#accuracy ↑MethodPaperTopic
186.1PaLI-X (55B)On Scaling Up a Multilingual Vision and Language Model
CVPR 2024 · Table 1
Visual question answering
282OFAOFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework
ICML 2022 · Table: cross-modal understandi
Cross-modal learning
381.15BridgeTower-large
CLIP-ViT-L-224/14 + RoBERTa-large
BridgeTower: Building Bridges between Encoders in Vision-Language Representation Learning
AAAI 2023 · Table: Comparisons with previo
Cross-modal learning
480.54METER-CoSwin_HUGE
CoSwin-Huge + RoBERTa-base
An Empirical Study of Training End-to-End Vision-and-Language Transformers
CVPR 2022 · Table 9
Human and embodied modeling
580.08UniTNT_BLIP
BLIP
Towards Models that Can See and Read
ICCV 2023 · Appendix SOTA table
Visual question answering
677.87Aurora (ours, r=64)
BLIP-base
Parameter-efficient Tuning of Large-scale Multimodal Foundation Model
NeurIPS 2023 · Table 3
CLIP and zero-shot
776.3HybridPromptHybridPrompt: Bridging Language Models and Human Priors in Prompt Tuning for Visual Question Answering
AAAI 2023 · Table 3
Visual question answering
875.8ToVE
CLIP ViT-B/16
ToVE: Efficient Vision-Language Learning via Knowledge Transfer from Vision Experts
ICLR 2025 · Table: Fine-tuned VQA performa
Multimodal LLMs
975.2IG captionerIG Captioner: Information Gain Captioners are Strong Zero-shot Classifiers
ECCV 2024 · Table 10
CLIP and zero-shot
1071.28MCAN + GSA
MCAN (Ours)
Normalized and Geometry-Aware Self-Attention Network for Image Captioning
CVPR 2020 · Table: Visual question answeri
Attention and transformers
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

VQAv2 · VQA score

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

#VQA score ↑MethodPaperTopic
183.9PaLI-X-VPD (55B)
PaLI-X (55B)
Visual Program Distillation: Distilling Tools and Programmatic Reasoning into Vision-Language Models
CVPR 2024 · Table 1
Multimodal LLMs
283.7CogAgent
CogVLM-17B
CogAgent: A Visual Language Model for GUI Agents
CVPR 2024 · VQA benchmark table
Multimodal LLMs
382.9CoS-7B
Vicuna
Accelerating Pre-training of Multimodal LLMs via Chain-of-Sight
NeurIPS 2024 · Comparison with existing appro
Multimodal LLMs
482.3CogVLM-Chat
Vicuna-7B
CogVLM: Visual Expert for Pretrained Language Models
NeurIPS 2024 · VQA and LVLM table
Human and embodied modeling
580.5LLaVA-1.5 + QA-ViT
Vicuna-7B
Question Aware Vision Transformer for Multimodal Reasoning
CVPR 2024 · Table 1
Human and embodied modeling
678.7Pink
Vicuna-7B
Pink: Unveiling the Power of Referential Comprehension for Multi-modal LLMs
CVPR 2024 · Table: conventional multi-moda
Multimodal LLMs
778.4AcFormer
Vicuna-7B
Visual Anchors Are Strong Information Aggregators For Multimodal Large Language Model
NeurIPS 2024 · General VQA results table
Multimodal LLMs
876.84InstructBLIP rel-att
InstructBLIP
MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMs
ICLR 2025 · Table: Accuracy of the propose
Visual question answering
969.77Qwen-VL-Chat LoRA-SFT on Ours-only
Qwen-VL-Chat
CertainlyUncertain: A Benchmark and Metric for Multimodal Epistemic and Aleatoric Awareness
ICLR 2025 · Table: Results on other benchm
Visual question answering
1061.9Img2LLM OPT 175B
OPT 175B
From Images to Textual Prompts: Zero-Shot Visual Question Answering With Frozen Large Language Models
CVPR 2023 · Table 2
Visual question answering

VQAv2 · overall accuracy (val)

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

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