| ORBIT: Overcoming Hallucination Risks via Bi-manifold Interaction and Traction | | 91.2 | ORBIT (Qwen2.5-VL-32B) | 2026-09-08 |
| Accelerating Multimodal Large Language Models with Prior-Corrected Token Reduction | ✓ Link | 90.8 | PriorTR (InternVL2.5-8B, 22.2% tokens) | 2026-06-23 |
| LoopVL: Recurrent Visual Intelligence | ✓ Link | 90.71 | LoopVL-1B | 2026-09-29 |
| Unlocking Dense Metric Depth Estimation in VLMs | ✓ Link | 89.9 | DepthVLM-4B | 2026-05-15 |
| MOSS-VL Technical Report | ✓ Link | 89.4 | MOSS-VL-Instruct-0708 | 2026-08-15 |
| Envision, Attend, Then Respond: Counterfactual Hallucination Mitigation in Large Vision-Language Models | ✓ Link | 89.0 | EnAR (InternVL3.5-8B) | 2026-06-03 |
| NeuPAT: Neuron-aware Plasticity Allocation Tuning for Language-Preserving MLLMs | | 88.72 | NeuPAT (Qwen3-4B-Instruct-2507) (extra data) | 2026-08-08 |
| From Pixels to Words -- Towards Native Vision-Language Primitives at Scale | ✓ Link | 88.4 | NEO-9B | 2025-10-16 |
| Grounded 3D-Aware Spatial Vision-Language Modeling | | 88.23 | GR3D-8B | 2026-05-28 |
| LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models | ✓ Link | 87.9 | LLaVA-NeXT-Interleave 14B | 2024-07-10 |
| MLLMs Hallucinate when Information Distribution Drifts in Synergy Heads | ✓ Link | 87.8 | HEAL (LLaVA-1.5-7B) | 2026-09-05 |
| LLaVA-NeXT: Improved reasoning, OCR, and world knowledge | ✓ Link | 87.73 | LLaVA-NeXT-34B | 2024-01-30 |
| Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision, Language, Audio, and Action | ✓ Link | 87.7 | Unified-IO 2 XXL | 2023-12-28 |
| Boosting Visual Instruction Tuning with Self-Supervised Guidance | ✓ Link | 87.7 | LLaVA-1.5-Qwen2.5-7B + V-GIFT | 2026-04-14 |
| An AI4AI Framework for Visual Token Pruning | | 87.67 | AutoPrune (LLaVA-1.5-7B, 32 visual tokens) | 2026-08-07 |
| VISion On Request: Enhanced VLLM efficiency with sparse, dynamically selected, vision-language interactions | | 87.6 | VISOR | 2026-03-24 |
| AnchorPrune: Relevance-Anchored Contextual Expansion for Visual Token Pruning | ✓ Link | 87.5 | LLaVA-NeXT-7B (AnchorPrune, 160 tokens) | 2026-07-08 |
| ViP-LLaVA: Making Large Multimodal Models Understand Arbitrary Visual Prompts | ✓ Link | 87.4 | ViP-LLaVA-13B | 2023-12-01 |
| Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling | ✓ Link | 87.4 | Janus-Pro-7B | 2025-01-29 |
| Switch-KD: Visual-Switch Knowledge Distillation for Vision-Language Models | ✓ Link | 87.3 | Switch-KD 0.5B | 2026-04-16 |
| ARM: An AutoRegressive Large Multimodal Model with Unified Discrete Representations | ✓ Link | 87.3 | ARM-7B | 2026-06-09 |
| MODUS: Decoder-Only Any-to-Any Modeling of Diverse Modalities | ✓ Link | 87.27 | Modus | 2026-07-28 |
| Token-level Response-visual Attention Guidance for Multimodal LLMs Knowledge Distillation | ✓ Link | 87.2 | TRAG (Qwen2.5-1.5B) | 2026-07-01 |
| Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation | ✓ Link | 87.0 | Janus | 2024-10-17 |
| Align-KD: Distilling Cross-Modal Alignment Knowledge for Mobile Vision-Language Large Model Enhancement | ✓ Link | 87.0 | MobileVLM V2 1.7B + Align-KD | 2024-12-02 |
| LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloud | | 86.93 | LFM2.5-VL-450M | 2026-04-08 |
| Script: Graph-Structured and Query-Conditioned Semantic Token Pruning for Multimodal Large Language Models | ✓ Link | 86.7 | LLaVA-NeXT-7B with Script | 2025-12-01 |
| Mixture of Specialized Vision Experts: Unlocking Complementary Visual Insights for Faithful MLLM Reasoning | ✓ Link | 86.5 | MoSVE | 2026-09-08 |
| Mobile-O: Unified Multimodal Understanding and Generation on Mobile Device | ✓ Link | 86.4 | Mobile-O-0.5B | 2026-02-23 |
| Online Self-Calibration Against Hallucination in Vision-Language Models | | 86.22 | LLaVA-1.5-7B (OSCAR) | 2026-05-01 |
| Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone | ✓ Link | 86.1 | Phi-3.5-Vision | 2024-04-22 |
| VisNec: Measuring and Leveraging Visual Necessity for Multimodal Instruction Tuning | ✓ Link | 86.0 | VisNec (LLaVA-v1.5-7B, LLaVA-665K 15%) | 2026-03-01 |
| Improved Baselines with Visual Instruction Tuning | ✓ Link | 85.9 | LLaVA-1.5-7B | 2023-10-05 |
| See Only When Needed: Context-Aware Attention Intervention for Hallucination-Free LVLMs | | 85.8 | CAI (LLaVA-1.5 7B) | 2026-09-08 |
| Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs | ✓ Link | 85.6 | Phi-4-Multimodal | 2025-02-24 |
| DICS: Exploring Data Intrinsic Consistency for Visual Instruction Selection | ✓ Link | 84.91 | DICS (LLaVA-1.5-7B, 25% LLaVA-665K) | 2026-08-31 |
| Video-LLaVA: Learning United Visual Representation by Alignment Before Projection | ✓ Link | 83.3 | Video-LLaVA-7B | 2023-11-16 |