| ERNIE 5.0 Technical Report | | 85.13 | ERNIE 5.0 | 2026-02-04 |
| Guidance Contrastive Token Credit Assignment for Discrete Policy Optimization | ✓ Link | 84.1 | Qwen3-VL-Instruct-8B + GCPO | 2026-05-28 |
| See Less, See Right: Bi-directional Perceptual Shaping For Multimodal Reasoning | ✓ Link | 77.4 | BiPS (Qwen3-VL-8B-Thinking) | 2025-12-26 |
| Self-Supervised Visual On-Policy Distillation | | 75.63 | S^2VOPD (Qwen3.5-9B, Vision-OPD-6K) (extra data) | 2026-08-14 |
| CodePercept: Code-Grounded Visual STEM Perception for MLLMs | ✓ Link | 73.56 | CodePercept (32B, S1) | 2026-03-11 |
| Learning to Select Visual Tools from Experience | | 73.1 | VisTA (GPT-o1) | 2026-06-03 |
| Beyond SFT-to-RL: Pre-alignment via Black-Box On-Policy Distillation for Multimodal RL | ✓ Link | 71.5 | PRISM+GSPO (Qwen3-VL-8B) | 2026-05-01 |
| LLaDA-o: An Effective and Length-Adaptive Omni Diffusion Model | ✓ Link | 71.1 | LLaDA-o | 2026-03-01 |
| Beyond Correctness: Benchmarking and Aligning Response Behaviors in Hybrid-Thinking MLLMs | | 70.21 | Qwen3-VL-4B + PatternRL | 2026-08-17 |
| Context-Aware RL for Agentic and Multimodal LLMs | ✓ Link | 66.4 | ContextRL (trained from Qwen3-VL-8B) | 2026-06-15 |
| OpenVLThinkerV2: A Generalist Multimodal Reasoning Model for Multi-domain Visual Tasks | ✓ Link | 65.8 | OpenVLThinkerV2 (Ours) | 2026-04-09 |
| Learning Consistency in Reward Modeling for Multi-Modal Reasoning | | 65.8 | CRM-RL-8B (Ours, soft-reward) | 2026-09-08 |
| EviRover: Reinforcing Agentic Perception Beyond a Glance | ✓ Link | 64.5 | EviRover | 2026-09-30 |
| OneThinker: All-in-one Reasoning Model for Image and Video | ✓ Link | 64.3 | OneThinker-8B | 2025-12-02 |
| Uncovering Understanding-Generation Synergy in Native Unified Multimodal Models: From Representation, Task to System | | 63.15 | Native UMM Synergy: Und+Gen geometry | 2026-09-01 |
| Learning Where Outcomes Change:Credit-Addressable Reasoning for Multimodal Geometry | | 62.44 | CE-GRPO (Qwen3-VL-8B) | 2026-08-31 |
| Insight-V++: Towards Advanced Long-Chain Visual Reasoning with Multimodal Large Language Models | | 62.4 | Insight-V++ | 2026-03-18 |
| AnE: Pushing the Reasoning Frontier of Multimodal LLMs via Anchor Evolution | ✓ Link | 62.3 | AnE-3rd | 2026-05-25 |
| Penguin-VL: Exploring the Efficiency Limits of VLM with LLM-based Vision Encoders | ✓ Link | 62.1 | Penguin-VL 8B | 2026-03-06 |
| Before Thinking, Learn to Decide: Proactive Routing for Efficient Visual Reasoning | | 61.2 | PRP (JRL, 7B) | 2026-06-29 |
| Near-Future Policy Optimization | | 60.0 | NPO (early + late-stage) | 2026-04-22 |
| Puzzle Curriculum GRPO for Vision-Centric Reasoning | ✓ Link | 58.1 | Qwen2.5-VL-7B (Mix) | 2025-12-16 |
| Dynamic Important Example Mining for Reinforcement Finetuning | ✓ Link | 58 | Qwen2.5-VL-32B + DIEM | 2026-08-29 |
| Youtu-VL: Unleashing Visual Potential via Unified Vision-Language Supervision | ✓ Link | 56.5 | Youtu-VL (4B) | 2026-01-27 |
| Beyond Where to Look: Trajectory-Guided Reinforcement Learning for Multimodal RLVR | | 55.21 | TGRL-DAPO (Qwen2.5-VL-7B, 2.1K) | 2026-03-27 |
| Thinking in Uncertainty: Mitigating Hallucinations in MLRMs with Latent Entropy-Aware Decoding | ✓ Link | 55.1 | VL-Cogito-7B with LEAD | 2026-03-09 |
| Not All Prompts Are Equal: Exploration-Guided Prompt Scaffolding for Multimodal Reinforcement Post-Training | | 55.1 | Qwen3-VL-4B + EPS (MMK12) | 2026-09-14 |
| V-Rubrics: Visual Faithfulness via Rubric-Based Reinforcement Learning | ✓ Link | 52.16 | V-Rubrics (+ GRPO) (extra data) | 2026-08-26 |
| Penguin-VL: Exploring the Efficiency Limits of VLM with LLM-based Vision Encoders | ✓ Link | 52.1 | Penguin-VL 2B | 2026-03-06 |
| Unlocking Token Rewards via Training-Free Reward Attribution | | 51.0 | Qwen2.5-VL-7B-Instruct + GRPO (P2T token reward) (extra data) | 2026-06-03 |
| Beyond Layer-Wise Merging: Chain-of-Merging for Vision-Language Models | | 50.25 | CoM (Qwen2.5-VL-7B + Qwen2.5-Math-7B) | 2026-06-03 |
| LLaDA-UI: Bringing Block-wise Diffusion to Vision-Language GUI Agents | ✓ Link | 48.0 | LLaDA-UI (pre-training) | 2026-09-09 |
| ReaGEN: Adaptive Generation of Structured Chains-of-Thought for Efficient Multimodal Reasoning | ✓ Link | 47.59 | ReaGEN (Qwen3-VL-4B, 4 iter) | 2026-06-03 |
| VideoAuto-R1: Video Auto Reasoning via Thinking Once, Answering Twice | ✓ Link | 46.9 | VideoAuto-R1 (Qwen2.5-VL-7B) | 2026-01-08 |
| CPPO: Contrastive Perception for Vision Language Policy Optimization | ✓ Link | 46.5 | CPPO (Qwen2.5-VL-7B) | 2026-01-01 |
| Process Rewards with Learned Reliability | ✓ Link | 46.4 | ACA with BetaPRM (Qwen2.5-VL-7B) | 2026-05-15 |
| PointThinker: Point-Incentivized Parallel Thinking for Multimodal Large Language Model | | 45.9 | PointThinker-7B (extra data) | 2026-06-03 |
| InternVL-U: Democratizing Unified Multimodal Models for Understanding, Reasoning, Generation and Editing | ✓ Link | 45.6 | InternVL-U (4B) | 2026-03-10 |
| AdaptVision: Efficient Vision-Language Models via Adaptive Visual Acquisition | ✓ Link | 42.3 | AdaptVision (Qwen2.5-VL-7B-Instruct) | 2025-12-03 |
| ARM-Thinker: Reinforcing Multimodal Generative Reward Models with Agentic Tool Use and Visual Reasoning | ✓ Link | 41.6 | ARM-Thinker-7B | 2025-12-04 |
| Training Data Efficiency in Multimodal Process Reward Models | ✓ Link | 40.89 | InternVL2.5-8B with MPRM (BIS-25%) | 2026-02-04 |
| Co-RL: Unsupervised Reasoning Emerges from Diverse Cohort in Multi-agent RL | ✓ Link | 40.74 | Co-RL (InternVL-3.5-8B, Different family) | 2026-08-19 |
| CogFlow: Bridging Perception and Reasoning through Knowledge Internalization for Visual Mathematical Problem Solving | ✓ Link | 39.5 | CogFlow (7B) | 2026-01-05 |
| Sparse-LaViDa: Sparse Multimodal Discrete Diffusion Language Models | | 37.9 | Sparse-LaViDa | 2025-12-16 |
| DiffusionVL: Translating Any Autoregressive Models into Diffusion Vision Language Models | ✓ Link | 36.29 | DiffusionVL | 2025-12-17 |
| Dream-VL & Dream-VLA: Open Vision-Language and Vision-Language-Action Models with Diffusion Language Model Backbone | ✓ Link | 31.5 | Dream-VL | 2025-12-27 |
| Reinforcement Learning for Multimodal Diffusion Language Models via Bidimensional Trajectory and Thought Optimization | | 31.2 | LLaDA-V + Bi-VRL (ours) | 2026-09-08 |