AI Paper Atlas

Research map / Benchmarks / BLINK

BLINK leaderboard

45 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on BLINK. Best accuracy: 85.3 (Qwen3-VL-4B-Instruct w/ MultihopSpatial-Train, ECCV 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.

BLINK · accuracy

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

#accuracy ↑MethodPaperTopic
185.3Qwen3-VL-4B-Instruct w/ MultihopSpatial-Train
Qwen3-VL-4B-Instruct
MultihopSpatial: Multi-hop Compositional Spatial Reasoning Benchmark for Vision-Language Model
ECCV 2026 · Table 2
Multimodal LLMs
281.82MetricReasonerMetric-Bench: Exploring In-context Spatial Metric Reasoning in VLMs for Indoor Scenes
ECCV 2026 · Table 4
Multimodal LLMs
381.8SpaceLLaVA + RoboSpatial
SpaceLLaVA
RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics
CVPR 2025 · Table: out-of-domain results
Multimodal LLMs
473.4SpatialClaw
Gemma4-31B
SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning
NeurIPS 2026 · Table 2 (left)
Multimodal LLMs
571L2-VMAS
Qwen3-VL-8B-Thinking
Dual Latent Memory for Visual Multi-agent System
ICML 2026 · Table 4
LLM agents
670.5VICO (Chart)
Qwen3-VL-8B
VICO: Visual Environments Co-Evolving for Vision-Language Model Reasoning
NeurIPS 2026 · Table 1
Multimodal LLMs
769.2VALOR
Qwen3-8B
No Labels, No Problem: Training Visual Reasoners with Multimodal Verifiers
ICLR 2026 · Table 1
Multimodal LLMs
866FactoSR-8B-RL
Qwen3-VL-8B-Instruct
Unfold The World: Factorize 4D Properties in Reinforcing Spatial Understanding
ECCV 2026 · Table 2
Multimodal LLMs
965.9OpenSpatial-Qwen2.5-VL-7B
Qwen2.5-VL-7B
OpenSpatial: A Principled Data Engine for Empowering Spatial Intelligence
ECCV 2026 · Table 1
Multimodal LLMs
1065.3CogSense-8B
Qwen3-VL-8B
Toward Cognitive Supersensing in Multimodal Large Language Models
NeurIPS 2026 · Table 2
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

BLINK · score

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

#score ↑MethodPaperTopic
168.5GeoSense
Qwen2.5-VL-3B + VGGT
GeoSense: Internalizing Geometric Necessity Perception for Multimodal Reasoning
ICML 2026 · Table 2
Multimodal LLMs
268.5Qwen2.5-VL-7B w. RIL
Qwen2.5-VL-7B
Unified Reinforcement and Imitation Learning for Vision-Language Models
NeurIPS 2025 · Table 1
Multimodal LLMs
358.53Qwen-PeRL
Qwen2.5-VL-7B-Instruct
PeRL: Permutation-Enhanced Reinforcement Learning for Interleaved Vision-Language Reasoning
NeurIPS 2025 · Main Results table
Multimodal LLMs
457.23MiCo-7B
Qwen2.5VL-7B
MiCo: Multi-image Contrast for Reinforcement Visual Reasoning
NeurIPS 2025 · Table: Quantitative results on
Multimodal LLMs
554.5Qwen2VL-7B-sft
Qwen2VL-7B
RAG-IGBench: Innovative Evaluation for RAG-based Interleaved Generation in Open-domain Question Answering
NeurIPS 2025 · Table: Performance of Qwen2VL
Multimodal LLMs
654.1Qwen2.5-VL +Cold Start +Multimodal RL
Qwen2.5-VL
Open Vision Reasoner: Transferring Linguistic Cognitive Behavior for Visual Reasoning
NeurIPS 2025 · Perception benchmark table
Multimodal LLMs
753.6Ours (RL)
Qwen2.5-VL-7B
Towards Policy-Adaptive Image Guardrail: Benchmark and Method
CVPR 2026 · Table 4
Multimodal LLMs
852.8UVU
Qwen2.5-3B-Instruct
UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm
CVPR 2026 · Table 4
Multimodal LLMs
947.7MIA-DPO
InternLM-XC2.5
MIA-DPO: Multi-Image Augmented Direct Preference Optimization For Large Vision-Language Models
ICLR 2025 · Main results on multi-image be
Multimodal LLMs
1043.5CoMemo
InternLM-1.8B
CoMemo: LVLMs Need Image Context with Image Memory
ICML 2025 · Table 2
Multimodal LLMs
1 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.