AI Paper Atlas

Research map / Benchmarks / GQA

GQA leaderboard

195 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on GQA. Best accuracy: 77 (R2-T2 (MoAI-7B, NGD), ICML 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.

GQA · accuracy

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

#accuracy ↑MethodPaperTopic
177R2-T2 (MoAI-7B, NGD)
MoAI-7B
R2-T2: Re-Routing in Test-Time for Multimodal Mixture-of-Experts
ICML 2025 · Table: comparison of three R2-
Human and embodied modeling
274.41SearchLVLMs with InternVL-1.5
InternVL-1.5
SearchLVLMs: A Plug-and-Play Framework for Augmenting Large Vision-Language Models by Searching Up-to-Date Internet Knowledge
NeurIPS 2024 · Table 2
Visual question answering
372.9CTGRCore-to-Global Reasoning for Compositional Visual Question Answering
AAAI 2025 · Table 1
Visual question answering
471.8CogSense-8B
Qwen3-VL-8B
Toward Cognitive Supersensing in Multimodal Large Language Models
NeurIPS 2026 · Table 2
Multimodal LLMs
570.51CFR+MLO
CFR
Consistency of Compositional Generalization Across Multiple Levels
AAAI 2025 · Table 2
Zero-/few-shot and meta-learning
669.5POLIA
Qwen2.5-VL-7B
POLIA: Policy Optimization with Visual-Object-Level Intrinsic Advantage for Multimodal Reasoning
ICML 2026 · Table 1
Multimodal LLMs
769.4Contrastive attention (ours)
LLaVA-1.5
Where Does Vision Meet Language? Understanding and Refining Visual Fusion in MLLMs via Contrastive Attention
CVPR 2026 · Table 1
Multimodal LLMs
869RandOpt
Qwen2.5-VL-3B-Inst
Neural Thickets: Diverse Task Experts Are Dense Around Pretrained Weights
ICML 2026 · Table 1
Mixture of experts
967.4Ours
Qwen2-VL-7B
Visual Enhanced Depth Scaling for Multimodal Latent Reasoning
NeurIPS 2026 · Table 1
Multimodal LLMs
1067GoM Text IDs + Rel labels
LlamaV-o1 11B
Graph-of-Mark: Promote Spatial Reasoning in Multimodal Language Models with Graph-Based Visual Prompting
AAAI 2026 · Table 2
Multimodal LLMs
106 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

GQA · score

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

#score ↑MethodPaperTopic
165.73MoCHA
Vicuna-7B
MoCHA: Advanced Vision-Language Reasoning with MoE Connector and Hierarchical Group Attention
AAAI 2026 · Table 1
Multimodal LLMs
265.7PaliGemma-2B (ViLex@224)
PaliGemma-2B
Visual Lexicon: Rich Image Features in Language Space
CVPR 2025 · Table 5
CLIP and zero-shot
365.3LLaVA-S² (Ours)
LLaVA-S², Vicuna-7B
DenseFusion-1M: Merging Vision Experts for Comprehensive Multimodal Perception
NeurIPS 2024 · Main results table
Multimodal LLMs
465.2FastVLM (R41)
Qwen2-7B
FastVLM: Efficient Vision Encoding for Vision Language Models
CVPR 2025 · VLM evaluations and comparison
Multimodal LLMs
564.9HAMSTER
VILA1.5-13B
HAMSTER: Hierarchical Action Models for Open-World Robot Manipulation
ICLR 2025 · Table: Core VQA and robustness
Robot action and navigation
664.8LEO
7B LLM
Rethinking the Mixture of Vision Encoders Paradigm for Enhanced Visual Understanding in Multimodal LLMs
NeurIPS 2026 · Table 4
Multimodal LLMs
764.6HaploVL-7B-Pro
Qwen2.5-7B
HaploVL: A Single-Transformer Baseline for Multi-Modal Understanding
ICML 2025 · Main results table
Language, safety and interpretability
864.6BASIC
Vicuna-13B
BASIC: Boosting Visual Alignment with Intrinsic Refined Embeddings in Multimodal Large Language Models
ICCV 2025 · Table 1
Multimodal LLMs
964.2CLOC in Open-LLaVA-NeXT
ViT-L/14; Vicuna-7B
Contrastive Localized Language-Image Pre-Training
ICML 2025 · Table: Multimodal benchmarks
CLIP and zero-shot
1063.9Mipha-3B
Phi-2 (2.7B)
A Comprehensive Overhaul of Multimodal Assistant with Small Language Models
AAAI 2025 · Table: Multi-modal evaluation
Multimodal LLMs
40 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

GQA · accuracy (test-dev)

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

#accuracy ↑MethodPaperTopic
177.35DACT (ponder cost 5e-3)
MAC
Differentiable Adaptive Computation Time for Visual Reasoning
CVPR 2020 · Supplementary Table 1
RNNs and normalization
265.05Oscar+ with VinVL (Base)
BERT-Base
VinVL: Revisiting Visual Representations in Vision-Language Models
CVPR 2021 · Table: overall comparison with
Open-set and few-shot detection
363.44GIVL
BERT-base
GIVL: Improving Geographical Inclusivity of Vision-Language Models With Pre-Training Methods
CVPR 2023 · Appendix Table 1
Human and embodied modeling
462.4D-ViTMDETR-LS+TP+TF
ViTMDETR
Dynamic Inference With Grounding Based Vision and Language Models
CVPR 2023 · Table 2
Human and embodied modeling
561.8BEiT-3+SUPS
BEiT-3
Compositional Substitutivity of Visual Reasoning for Visual Question Answering
ECCV 2024 · Table 3
Visual question answering
661.42CLIP-ViLp CLIP-Res50x4
BERT-Base
How Much Can CLIP Benefit Vision-and-Language Tasks?
ICLR 2022 · Table 6
CLIP and zero-shot
760.9Ours, VinVL
VL-Transformer
Supervising the Transfer of Reasoning Patterns in VQA
NeurIPS 2021 · Table 3
Visual question answering
860.1UniPT
CLIP-ViL
UniPT: Universal Parallel Tuning for Transfer Learning with Efficient Parameter and Memory
CVPR 2024 · Table 3
Pruning and efficient tuning
959.7MoME
Vicuna-v1.5 (7B)
MoME: Mixture of Multimodal Experts for Generalist Multimodal Large Language Models
NeurIPS 2024 · Table: Comparison with state-o
Multimodal LLMs
1059.6Lite-MDETR-TTQ
MDETR-ENB5
Lite-MDETR: A Lightweight Multi-Modal Detector
CVPR 2022 · Table 2
CNN features
13 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

GQA · accuracy (val)

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

2 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.