TraceAV-Bench is the first benchmark for multi-hop trajectory reasoning over long audio-visual videos, showing top models reach only 51-68% accuracy with substantial room for improvement.
super hub
InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models
140 Pith papers cite this work. Polarity classification is still indexing.
140
Pith papers citing it
abstract
We introduce InternVL3, a significant advancement in the InternVL series featuring a native multimodal pre-training paradigm. Rather than adapting a text-only large language model (LLM) into a multimodal large language model (MLLM) that supports visual inputs, InternVL3 jointly acquires multimodal and linguistic capabilities from both diverse multimodal data and pure-text corpora during a single pre-training stage. This unified training paradigm effectively addresses the complexities and alignment challenges commonly encountered in conventional post-hoc training pipelines for MLLMs. To further improve performance and scalability, InternVL3 incorporates variable visual position encoding (V2PE) to support extended multimodal contexts, employs advanced post-training techniques such as supervised fine-tuning (SFT) and mixed preference optimization (MPO), and adopts test-time scaling strategies alongside an optimized training infrastructure. Extensive empirical evaluations demonstrate that InternVL3 delivers superior performance across a wide range of multi-modal tasks. In particular, InternVL3-78B achieves a score of 72.2 on the MMMU benchmark, setting a new state-of-the-art among open-source MLLMs. Its capabilities remain highly competitive with leading proprietary models, including ChatGPT-4o, Claude 3.5 Sonnet, and Gemini 2.5 Pro, while also maintaining strong pure-language proficiency. In pursuit of open-science principles, we will publicly release both the training data and model weights to foster further research and development in next-generation MLLMs.
hub tools
citation-role summary
baseline 1
citation-polarity summary
claims ledger
- abstract We introduce InternVL3, a significant advancement in the InternVL series featuring a native multimodal pre-training paradigm. Rather than adapting a text-only large language model (LLM) into a multimodal large language model (MLLM) that supports visual inputs, InternVL3 jointly acquires multimodal and linguistic capabilities from both diverse multimodal data and pure-text corpora during a single pre-training stage. This unified training paradigm effectively addresses the complexities and alignment challenges commonly encountered in conventional post-hoc training pipelines for MLLMs. To further
authors
co-cited works
roles
baseline 1polarities
baseline 1representative citing papers
MedHorizon benchmark reveals current multimodal LLMs achieve only 41.1% accuracy on long medical videos due to failures in sparse evidence retrieval and procedural reasoning.
S1-VL combines structured scientific reasoning with iterative image manipulation via code execution to reach state-of-the-art results on visual and scientific reasoning benchmarks.
EVE enables verifiable self-evolution of MLLMs by using a Challenger-Solver architecture to generate dynamic executable visual transformations that produce VQA problems with absolute execution-verified ground truth.
VLM-UnBench demonstrates that prompt-based training-free unlearning in VLMs leaves forget accuracy near the no-instruction baseline except under oracle conditions that reveal the target concept.
VLMs over-correct multi-line handwritten math OCR, and the PINK metric using LLM rubric grading penalizes this for better human alignment.
G²TR reduces visual tokens and prefill computation by 1.94x in separate-encoder UMMs via generation-guided importance from VAE latent consistency while preserving reasoning accuracy and editing quality.
PRISM-VL improves VLM performance by grounding on RAW-derived Meas.-XYZ inputs and exposure-bracketed supervision, gaining +0.1074 BLEU and +4.46% LLM-Judge accuracy over an RGB baseline on a held-out benchmark.
Multi-grained counting is introduced with five granularity levels, supported by the new KubriCount dataset generated via 3D synthesis and editing, and HieraCount model that combines text and visual exemplars for improved accuracy.
BenchCAD benchmark shows frontier multimodal models recover coarse geometry but fail to produce accurate parametric CAD programs for industrial parts, with limited generalization after fine-tuning.
TOC-Bench is a new diagnostic benchmark that reveals major weaknesses in temporal object consistency for Video-LLMs, including event counting, ordering, identity reasoning, and hallucination avoidance.
State-conditioned commitment depth in a vision-language policy Pareto-dominates fixed-depth baselines on Sliding Puzzle and Sokoban, raising solve rates by up to 12.5 points while using 25% fewer actions and beating larger models.
STEMO-Bench evaluates intermediate spatio-temporal reasoning in video MLLMs via object-centric facts, and STEMO-Track improves consistency by chunk-wise trajectory construction and aggregation.
SphereVAD performs training-free video anomaly detection by recasting anomaly discrimination as von Mises-Fisher likelihood-ratio geodesic inference on the unit hypersphere using intermediate MLLM features, with Frechet mean centering, holistic scene attention, and spherical geodesic pulling.
GazeVLM introduces internal gaze tokens that allow VLMs to dynamically suppress irrelevant visual features and simulate foveal attention for improved high-resolution multimodal reasoning.
PolarVLM is the first VLM framework to integrate polarimetric physical parameters via dual-stream architecture and progressive training, delivering 25.4% gains over RGB baselines on reflection and transparency tasks with a new 75K-pair PolarVQA benchmark.
ScaleEarth conditions remote sensing VLMs on continuous GSD via CS-HLoRA and a visual GSD predictor, creating a closed training loop with GeoScale-VQA to achieve SOTA on Earth observation benchmarks.
SRPO refines GRPO into role-aware token-level advantages by emphasizing perception tokens based on visual dependency (original vs. corrupted inputs) and reasoning tokens based on consistency with perception, unified via a shared baseline.
Pest-Thinker is a reinforcement learning framework that improves MLLMs' expert-level reasoning on pest morphology via synthesized CoT trajectories, GRPO optimization, and an LLM-judged feature reward on new benchmarks QFSD and AgriInsect.
VT-Bench is the first unified benchmark aggregating 14 visual-tabular datasets with over 756K samples and evaluating 23 models to expose challenges in this multi-modal area.
GEASS selectively gates and weights self-generated captions using confidence and entropy to reduce object hallucinations in VLMs, outperforming vanilla inference and contrastive decoding on POPE and HallusionBench.
Introduces QCalEval benchmark showing best zero-shot VLM score of 72.3 on quantum calibration plots, with fine-tuning and in-context learning effects varying by model type.
FCMBench-Video is a new benchmark with 1,200 videos and 11k QA instances for evaluating Video-MLLMs on document video understanding across 28 document types.
Perceval is a perception-centric PRM that detects token-level perceptual errors in VLMs, supporting token-advantage RL training and iterative test-time scaling for improved reasoning.
citing papers explorer
-
TraceAV-Bench: Benchmarking Multi-Hop Trajectory Reasoning over Long Audio-Visual Videos
TraceAV-Bench is the first benchmark for multi-hop trajectory reasoning over long audio-visual videos, showing top models reach only 51-68% accuracy with substantial room for improvement.
-
MedHorizon: Towards Long-context Medical Video Understanding in the Wild
MedHorizon benchmark reveals current multimodal LLMs achieve only 41.1% accuracy on long medical videos due to failures in sparse evidence retrieval and procedural reasoning.
-
S1-VL: Scientific Multimodal Reasoning Model with Thinking-with-Images
S1-VL combines structured scientific reasoning with iterative image manipulation via code execution to reach state-of-the-art results on visual and scientific reasoning benchmarks.
-
EVE: Verifiable Self-Evolution of MLLMs via Executable Visual Transformations
EVE enables verifiable self-evolution of MLLMs by using a Challenger-Solver architecture to generate dynamic executable visual transformations that produce VQA problems with absolute execution-verified ground truth.
-
Can VLMs Truly Forget? Benchmarking Training-Free Visual Concept Unlearning
VLM-UnBench demonstrates that prompt-based training-free unlearning in VLMs leaves forget accuracy near the no-instruction baseline except under oracle conditions that reveal the target concept.
-
When VLMs 'Fix' Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR
VLMs over-correct multi-line handwritten math OCR, and the PINK metric using LLM rubric grading penalizes this for better human alignment.
-
G$^2$TR: Generation-Guided Visual Token Reduction for Separate-Encoder Unified Multimodal Models
G²TR reduces visual tokens and prefill computation by 1.94x in separate-encoder UMMs via generation-guided importance from VAE latent consistency while preserving reasoning accuracy and editing quality.
-
Allegory of the Cave: Measurement-Grounded Vision-Language Learning
PRISM-VL improves VLM performance by grounding on RAW-derived Meas.-XYZ inputs and exposure-bracketed supervision, gaining +0.1074 BLEU and +4.46% LLM-Judge accuracy over an RGB baseline on a held-out benchmark.
-
Count Anything at Any Granularity
Multi-grained counting is introduced with five granularity levels, supported by the new KubriCount dataset generated via 3D synthesis and editing, and HieraCount model that combines text and visual exemplars for improved accuracy.
-
BenchCAD: A Comprehensive, Industry-Standard Benchmark for Programmatic CAD
BenchCAD benchmark shows frontier multimodal models recover coarse geometry but fail to produce accurate parametric CAD programs for industrial parts, with limited generalization after fine-tuning.
-
TOC-Bench: A Temporal Object Consistency Benchmark for Video Large Language Models
TOC-Bench is a new diagnostic benchmark that reveals major weaknesses in temporal object consistency for Video-LLMs, including event counting, ordering, identity reasoning, and hallucination avoidance.
-
When to Re-Commit: Temporal Abstraction Discovery for Long-Horizon Vision-Language Reasoning
State-conditioned commitment depth in a vision-language policy Pareto-dominates fixed-depth baselines on Sliding Puzzle and Sokoban, raising solve rates by up to 12.5 points while using 25% fewer actions and beating larger models.
-
Tracking the Truth: Object-Centric Spatio-Temporal Monitoring for Video Large Language Models
STEMO-Bench evaluates intermediate spatio-temporal reasoning in video MLLMs via object-centric facts, and STEMO-Track improves consistency by chunk-wise trajectory construction and aggregation.
-
SphereVAD: Training-Free Video Anomaly Detection via Geodesic Inference on the Unit Hypersphere
SphereVAD performs training-free video anomaly detection by recasting anomaly discrimination as von Mises-Fisher likelihood-ratio geodesic inference on the unit hypersphere using intermediate MLLM features, with Frechet mean centering, holistic scene attention, and spherical geodesic pulling.
-
GazeVLM: Active Vision via Internal Attention Control for Multimodal Reasoning
GazeVLM introduces internal gaze tokens that allow VLMs to dynamically suppress irrelevant visual features and simulate foveal attention for improved high-resolution multimodal reasoning.
-
PolarVLM: Bridging the Semantic-Physical Gap in Vision-Language Models
PolarVLM is the first VLM framework to integrate polarimetric physical parameters via dual-stream architecture and progressive training, delivering 25.4% gains over RGB baselines on reflection and transparency tasks with a new 75K-pair PolarVQA benchmark.
-
Beyond GSD-as-Token: Continuous Scale Conditioning for Remote Sensing VLMs
ScaleEarth conditions remote sensing VLMs on continuous GSD via CS-HLoRA and a visual GSD predictor, creating a closed training loop with GeoScale-VQA to achieve SOTA on Earth observation benchmarks.
-
Structured Role-Aware Policy Optimization for Multimodal Reasoning
SRPO refines GRPO into role-aware token-level advantages by emphasizing perception tokens based on visual dependency (original vs. corrupted inputs) and reasoning tokens based on consistency with perception, unified via a shared baseline.
-
Pest-Thinker: Learning to Think and Reason like Entomologists via Reinforcement Learning
Pest-Thinker is a reinforcement learning framework that improves MLLMs' expert-level reasoning on pest morphology via synthesized CoT trajectories, GRPO optimization, and an LLM-judged feature reward on new benchmarks QFSD and AgriInsect.
-
VT-Bench: A Unified Benchmark for Visual-Tabular Multi-Modal Learning
VT-Bench is the first unified benchmark aggregating 14 visual-tabular datasets with over 756K samples and evaluating 23 models to expose challenges in this multi-modal area.
-
GEASS: Training-Free Caption Steering for Hallucination Mitigation in Vision-Language Models
GEASS selectively gates and weights self-generated captions using confidence and entropy to reduce object hallucinations in VLMs, outperforming vanilla inference and contrastive decoding on POPE and HallusionBench.
-
QCalEval: Benchmarking Vision-Language Models for Quantum Calibration Plot Understanding
Introduces QCalEval benchmark showing best zero-shot VLM score of 72.3 on quantum calibration plots, with fine-tuning and in-context learning effects varying by model type.
-
FCMBench-Video: Benchmarking Document Video Intelligence
FCMBench-Video is a new benchmark with 1,200 videos and 11k QA instances for evaluating Video-MLLMs on document video understanding across 28 document types.
-
Improving Vision-language Models with Perception-centric Process Reward Models
Perceval is a perception-centric PRM that detects token-level perceptual errors in VLMs, supporting token-advantage RL training and iterative test-time scaling for improved reasoning.
-
CGC: Compositional Grounded Contrast for Fine-Grained Multi-Image Understanding
CGC improves fine-grained multi-image understanding in MLLMs by constructing contrastive training instances from existing single-image annotations and adding a rule-based spatial reward, achieving SOTA on MIG-Bench and VLM2-Bench with transfer gains to other multimodal tasks.
-
SpaMEM: Benchmarking Dynamic Spatial Reasoning via Perception-Memory Integration in Embodied Environments
SpaMEM benchmark shows multimodal LLMs succeed at spatial tasks with text histories but sharply fail at long-horizon belief maintenance from raw visual streams alone.
-
Divide-then-Diagnose: Weaving Clinician-Inspired Contexts for Ultra-Long Capsule Endoscopy Videos
Introduces the diagnosis-driven CE video summarization task, the VideoCAP dataset with 240 annotated videos, and the DiCE framework that outperforms prior methods by screening candidates then weaving them into diagnostic contexts.
-
X-PCR: A Benchmark for Cross-modality Progressive Clinical Reasoning in Ophthalmic Diagnosis
X-PCR is a new benchmark of 26,415 images and 177,868 expert VQA pairs that evaluates MLLMs on six-stage progressive reasoning and cross-modality integration in ophthalmology.
-
Hybrid Latent Reasoning with Decoupled Policy Optimization
HyLaR with DePO enables effective RL in hybrid discrete-continuous spaces for multimodal models, outperforming prior MLLMs on perception and understanding benchmarks.
-
Culture-Aware Humorous Captioning: Multimodal Humor Generation across Cultural Contexts
Introduces culture-aware humorous captioning task and staged alignment framework that improves contextual fit and balances image relevance with humor in multimodal LLMs.
-
GaLa: Hypergraph-Guided Visual Language Models for Procedural Planning
GaLa uses hypergraph representations of objects and a TriView encoder with contrastive learning to improve vision-language models on procedural planning benchmarks.
-
S-GRPO: Unified Post-Training for Large Vision-Language Models
S-GRPO unifies SFT and RL for LVLMs via conditional ground-truth injection that supplies a maximal-reward anchor when group exploration fails completely.
-
HyperGVL: Benchmarking and Improving Large Vision-Language Models in Hypergraph Understanding and Reasoning
HyperGVL is the first benchmark for LVLMs on hypergraph tasks from basic counting to NP-hard reasoning, with 12 models tested and a router proposed to adapt representations.
-
MCSC-Bench: Multimodal Context-to-Script Creation for Realistic Video Production
MCSC-Bench is the first large-scale dataset for the Multimodal Context-to-Script Creation task, requiring models to select relevant shots from redundant materials, plan missing shots, and generate coherent scripts with voiceovers.
-
Why MLLMs Struggle to Determine Object Orientations
Orientation information is recoverable from MLLM visual encoder embeddings via linear regression, contradicting the hypothesis that failures originate in the encoders.
-
Semantic-Geometric Dual Compression: Training-Free Visual Token Reduction for Ultra-High-Resolution Remote Sensing Understanding
DualComp uses a lightweight router to split visual token compression into a semantic stream with size-adaptive clustering and a geometric stream with path-tracing recovery, enabling low-cost high-fidelity UHR remote sensing interpretation.
-
OmniScript: Towards Audio-Visual Script Generation for Long-Form Cinematic Video
OmniScript is a new 8B omni-modal model that turns long cinematic videos into scene-by-scene scripts and matches top proprietary models on temporal localization and semantic accuracy.
-
MMR-AD: A Large-Scale Multimodal Dataset for Benchmarking General Anomaly Detection with Multimodal Large Language Models
MMR-AD is a new benchmark dataset showing that current generalist MLLMs lag industrial needs for anomaly detection, with Anomaly-R1 delivering better results through reasoning and RL.
-
EgoTL: Egocentric Think-Aloud Chains for Long-Horizon Tasks
EgoTL provides a new egocentric dataset with think-aloud chains and metric labels that benchmarks VLMs on long-horizon tasks and improves their planning, reasoning, and spatial grounding after finetuning.
-
ActFER: Agentic Facial Expression Recognition via Active Tool-Augmented Visual Reasoning
ActFER reformulates facial expression recognition as active tool-augmented visual reasoning with a custom reinforcement learning algorithm UC-GRPO that outperforms passive MLLM baselines on AU prediction.
-
CrashSight: A Phase-Aware, Infrastructure-Centric Video Benchmark for Traffic Crash Scene Understanding and Reasoning
CrashSight is a new infrastructure-focused benchmark showing that state-of-the-art vision-language models can describe crash scenes but fail at temporal and causal reasoning.
-
ViVa: A Video-Generative Value Model for Robot Reinforcement Learning
ViVa turns a video generator into a value model for robot RL that jointly forecasts future states and task value, yielding better performance on real-world box assembly when integrated with RECAP.
-
Bridging Time and Space: Decoupled Spatio-Temporal Alignment for Video Grounding
Bridge-STG decouples spatio-temporal alignment via semantic bridging and query-guided localization modules to achieve state-of-the-art m_vIoU of 34.3 on VidSTG among MLLM methods.
-
MARINER: A 3E-Driven Benchmark for Fine-Grained Perception and Complex Reasoning in Open-Water Environments
MARINER is a new benchmark dataset and evaluation framework for fine-grained perception and causal reasoning in open-water scenes using 16,629 images across 63 vessel categories, diverse environments, and maritime incidents.
-
DISSECT: Diagnosing Where Vision Ends and Language Priors Begin in Scientific VLMs
DISSECT benchmark reveals that VLMs extract visual details from scientific diagrams but frequently lose them during reasoning, with open-source models showing a larger integration gap than closed-source ones.
-
The Blind Spot of Adaptation: Quantifying and Mitigating Forgetting in Fine-tuned Driving Models
Fine-tuning VLMs for driving erodes pre-trained world knowledge, but shifting adaptation to prompt space via the Drive Expert Adapter preserves generalization while improving task performance.
-
TableVision: A Large-Scale Benchmark for Spatially Grounded Reasoning over Complex Hierarchical Tables
TableVision benchmark shows explicit spatial grounding recovers MLLM reasoning on hierarchical tables, delivering 12.3% accuracy improvement through a decoupled perception-reasoning framework.
-
Token Warping Helps MLLMs Look from Nearby Viewpoints
Backward token warping in ViT-based MLLMs enables reliable reasoning from nearby viewpoints by preserving semantic coherence better than pixel-wise warping or fine-tuning baselines.
-
V-Reflection: Transforming MLLMs from Passive Observers to Active Interrogators
V-Reflection introduces a think-then-look mechanism where MLLM latent states actively interrogate visual features via two-stage distillation from a box-guided teacher to a dynamic autoregressive student, narrowing the fine-grained perception gap on benchmarks.
-
DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning
DeepEyes uses reinforcement learning to teach vision-language models active perception and image-based thinking, yielding gains on perception, reasoning, grounding, and hallucination benchmarks.