The DOI matches its Crossref record; 94 of the 110 references it cites were found in a registry or an index.
OCRBench: on the hidden mystery of OCR in large multimodal models
Open accessOverview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“benchmarks targeting a single perceptual application, such as OCRBench (Yuliang Liu et al. 2024) and ScreenSpot (K. Li et al. 2025), provide more direct measures of perception but are built top-down: their categories are predefined and populated with questions from a specific application domain.”
From PerceptionBench: Evaluating Atomic Visual Perception in Multimodal Large Language Models (2026) · cites this paper as Yuliang Liu et al. 2024
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Learning Transferable Visual Models From Natural Language Supervision
Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. (2021) · arXiv · doi:10.48550/arxiv.2103.00020 · Cited by 15,293 · cited 1×
“The success of large models has also been extended to the multimodal vision-language space [7], leading to a line of research on large multimodal models (LMM), including contrastive learning [8, 9, 10, 11] and generative modeling [2, 12, 13, 14, 15].”
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., & Liu, P. J. (2019) · arXiv · doi:10.48550/arxiv.1910.10683 · Cited by 10,929
Hawkins, C., Sorgi, M., & Lock, S. (1985) · Springer London · doi:10.1007/978-1-4471-3519-7 · Cited by 9,160 · cited 1×
“5 [64] Shanghai AI Lab 720 DeepSeek-VL-7B [81] DeepSeek 435 Claude3-Opus [82] Anthropic 694 OmniLMM-12B [83] OpenBMB 420 RekaFlash [74] Reka AI 692 TransCore-M [84] PCI Research 405 InternLM-XComposer2.”
LLaMA: Open and Efficient Foundation Language Models
Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.13971 · Cited by 4,602 · cited 1×
“Among them, LLaMA [3] is an open-source LLM that matches the performance of GPT-3, followed by Alpaca [4], Vicuna [5], GPT-4-LLM [6] to improve the LLM’s alignment ability to follow human instruction, reporting impressive performance compared with proprietary LLMs.”
Liu, H., Li, C., Wu, Q., & Lee, Y. J. (2023) · Advances in Neural Information Processing Systems 36 · doi:10.52202/075280-1516 · Cited by 2,428
Self-citationChartQA: A Benchmark for Question Answering about Charts with Visual and Logical ReasoningPreprint
Masry, A., Long, D. X., Tan, J., Joty, S. R., & Hoque, E. (2022) · Findings · doi:10.48550/arxiv.2203.10244 · Cited by 2,023 · cited 1×
“We assess these LMMs on DocVQA[55], InfographicVQA [56] and ChartQA [57].”
Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision
Jia, C., Yang, Y., Xia, Y., Chen, Y.-T., Parekh, Z., Pham, H., Le, Q. V., Sung, Y., Li, Z., & Duerig, T. (2021) · arXiv · doi:10.48550/arxiv.2102.05918 · Cited by 1,977 · cited 1×
“The success of large models has also been extended to the multimodal vision-language space [7], leading to a line of research on large multimodal models (LMM), including contrastive learning [8, 9, 10, 11] and generative modeling [2, 12, 13, 14, 15].”
Self-citationImproved Baselines with Visual Instruction Tuning
Liu, H., Li, C., Li, Y., & Lee, Y. J. (2024) · 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52733.2024.02484 · Cited by 1,748 · cited 1×
“Other works, such as [20, 21], also focus on aligning the vision module and LLM for improved multimodal understanding.”
Self-citationToward expert-level medical question answering with large language models
Singhal, K., Tu, T., Gottweis, J., Sayres, R., Wulczyn, E., Amin, M., Hou, L., Clark, K., Pfohl, S. R., Cole-Lewis, H., et al. (2025) · Nature Medicine · doi:10.1038/s41591-024-03423-7 · Cited by 1,734
Self-citationICDAR 2015 competition on Robust Reading
Karatzas, D., Gomez-Bigorda, L., Nicolaou, A., Ghosh, S. K., Bagdanov, A. D., Iwamura, M., Matas, J., Neumann, L., Chandrasekhar, V., Lu, S.-J., et al. (2015) · IEEE International Conference on Document Analysis and Recognition · doi:10.1109/icdar.2015.7333942 · Cited by 1,622 · cited 2×
“…OCR text recognition datasets, including (1) Regular Text Recognition: IIIT5K [37], SVT [38], IC13 [39]; (2)Irregular Text Recognition: IC15 [40], SVTP [41], CT80 [42], COCOText(COCO) [43], SCUT-CTW1500 (CTW) [44], Total-Text (TT) [45]; (3) Occlusion Scene Text [46], encompassing weakly…”
Flagged first, then by citations
16 of the 110 references couldn’t be matched to a known paper.
Research works printed but not linked: 14
- Achiam J, Adler S, Agarwal S, et al. GPT-4 technical report. 2023. ArXiv:2303.08774 · arXiv
- Vicuna. Vicuna: an open-source chatbot impressing GPT-4 with 90%* ChatGPT quality. 2023. https://vicuna.lmsys.org.
- Google Gemini Team. Gemini: a family of highly capable multimodal models. 2023. ArXiv:2312.11805 · arXiv
- OpenAI. GPT-4V(vision) system card. 2023. https://openai.com/index/gpt-4v-system-card
- Park S, Shin S, Lee B, et al. CORD: a consolidated receipt dataset for post-ocr parsing. In: Proceedings of the Document Intelligence Workshop at Neural Information Processing Systems, 2019
- CloudWalk. Congrong. 2024. https://www.cloudwalk.com
- Huang M X, Liu Y L, Liang D K, et al. Mini-Monkey: alleviate the sawtooth effect by multi-scale adaptive cropping. 2024. ArXiv:2408.02034 · arXiv
- OpenAI. GPT-4o-mini-20240718. 2024. https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence
- Reka AI. Rekaflash. 2024. https://www.reka.ai
- Google. Gemini models. 2024. https://deepmind.google/technologies/gemini
- Lu H Y, Liu W, Zhang B, et al. DeepSeek-VL: towards real-world vision-language understanding. 2024. ArXiv:2403.05525 · arXiv
- Su Y X, Lan T, Li H Y, et al. PandaGPT: one model to instruction-follow them all. 2023. ArXiv:2305.16355 · arXiv
- StepFun. Step-1v. https://platform.stepfun.com, 2024
- Team Chameleon. Chameleon: mixed-modal early-fusion foundation models. 2024. ArXiv:2405.09818 · arXiv
Cited by
Sorted by citations
Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAsPreprint
Microsoft, :, Abouelenin, A., Ashfaq, A., Atkinson, A., Awadalla, H. H., Bách, N., Bao, J., Benhaim, A., Cai, M., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.01743 · Cited by 669
Self-citationMMMU: A Massive Multi-Discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI
Yue, X., Ni, Y., Zheng, T., Zhang, K., Liu, R., Zhang, G., Stevens, S., Jiang, D., Ren, W., Sun, Y., et al. (2024) · 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52733.2024.00913 · Cited by 429
Self-citationLLaVA-OneVision-1.5: Fully Open Framework for Democratized Multimodal TrainingPreprint
An, X., Xie, Y., Yang, K., Zhang, W., Zhao, X., Cheng, Z., Wang, Y., Xu, S., Chen, C., Zhu, D., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.23661 · Cited by 302
“5 RL (Fast) Qwen 2.5-VL”
InfluentialSelf-citationV Team, Hong, W., Yu, W., Gu, X., Wang, G., Gan, G., Tang, H., Cheng, J., Qi, J., Ji, J., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.01006 · Cited by 606
“From solving scientific problems [67; 32; 34] to developing autonomous agents [21; 62; 43], the demands on VLMs have far surpassed simple visual content perception [31], with an increasing emphasis on advanced reasoning abilities.”
Self-citationTextMonkey: An OCR-Free Large Multimodal Model for Understanding Document
Liu, Y., Yang, B., Liu, Q., Li, Z., Ma, Z., Zhang, S., & Bai, X. (2026) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2026.3653415 · Cited by 103
“The selection of test datasets and the formulation of evaluation criteria were carried out in accordance with the methodology described in [10].”
InfluentialSelf-citationRoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete
Ji, Y., Tan, H., Shi, J., Hao, X., Zhang, Y., Zhang, H., Wang, P., Zhao, M., Mu, Y., An, P., et al. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.00168 · Cited by 55
TokenPacker: Efficient Visual Projector for Multimodal LLM
Li, W., Yuan, Y., Liu, J., Tang, D., Wang, S., Qin, J., Zhu, J., & Zhang, L. (2025) · International Journal of Computer Vision · doi:10.1007/s11263-025-02491-7 · Cited by 159
MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe
Yu, T.-Y., Wang, Z., Wang, C., Huang, F.-W., Ma, W., He, Z., Cai, T., Chen, W., Huang, Y.-X., Zhao, Y., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2509.18154 · Cited by 211
“[53], ChartQA [54], TextVQA [55], DocVQA [56], and OmniDocBench [57], testing ability to extract, interpret, and reason about textual information in various visual contexts, including documents and charts.”
Qwen3.5-Omni Technical ReportPreprint
Qwen Team (2026) · arXiv.org · doi:10.48550/arxiv.2604.15804 · Cited by 210
“For the general visual question answering, the model is evaluated on RealWorldQA (Zhang et al., 2024), MMStar (Chen et al., 2024a), HallusionBench (Guan et al., 2024), and SimpleVQA (Cheng et al., 2025).”
Cambrian-S: Towards Spatial Supersensing in Video
Yang, S.-S., Yang, J.-H., Huang, P.-Z., Brown, E., Yang, Z., Yu, Y., Tong, S.-B., Zheng, Z.-H., Xu, Y.-F., Wang, M.-H., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2511.04670 · Cited by 167
Efficient multimodal large language models: a survey
Jin, Y., Li, J., Gu, T., Liu, Y., Zhao, B., Lai, J., Gan, Z., Wang, Y., Wang, C., Tan, X., et al. (2025) · Visual Intelligence · doi:10.1007/s44267-025-00099-6 · Cited by 184
“For tasks requiring fine-grained text recognition, OCR-Bench [205] provides a robust testbed for evaluating optical character recognition capabilities in diverse scenes.”
Self-citationZhang, Y.-F., Lu, X., Yin, S.-K., Fu, C.-Y., Chen, W., Hu, X., Wen, B., Jiang, K.-Y., Liu, C.-Y., Zhang, T.-K., et al. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2508.11630 · Cited by 159
“These benchmarks include Hallucination bench (Li et al., 2023), MMStar (Chen et al., 2024), MMVet Hard (Yu et al., 2024), OCR Bench (Liu et al., 2023), Chart QA (Masry et al., 2022), and BLINK (Fu et al., 2024a).”
Self-citationFrom Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D
Zhang, J.-H., Chen, Y., Zhou, Y., Xu, Y., Huang, Z., Mei, J.-L., Chen, J., Yuan, Y.-J., Cai, X., Huang, G., et al. (2025) · Neural Information Processing Systems · doi:10.48550/arxiv.2503.22976 · Cited by 153
“Existing multi-modal datasets primarily provide single-view or limited multi-view images, often designed for image-text retrieval or basic visual question answering, making them insufficient for advanced 3D reasoning [33, 38, 39, 54].”
ChatVLA: Unified Multimodal Understanding and Robot Control with Vision-Language-Action Model
Zhou, Z.-Y., Zhu, Y.-C., Zhu, M., Wen, J.-J., Liu, N., Xu, Z.-Y., Meng, W.-B., Cheng, R., Peng, Y.-X., Shen, C.-M., et al. (2025) · Conference on Empirical Methods in Natural Language Processing · doi:10.48550/arxiv.2502.14420 · Cited by 151
“We also tested against more challenging benchmarks designed for MLLMs, i.e., MMMU [62], MMStar [9], MME [19], OCR-Bench [35], HallBench [20] and MMBench [34].”
MMSearch-R1: Incentivizing LMMs to Search
Wu, J., Deng, Z.-H., Li, W., Liu, Y., You, B., Li, B., Ma, Z.-J., & Liu, Z. (2025) · Annual Meeting of the Association for Computational Linguistics · doi:10.48550/arxiv.2506.20670 · Cited by 137
Self-citationBLIP-3: A Family of Open Large Multimodal Models
Xue, L., Shu, M., Awadalla, A., Wang, J., Yan, A., Purushwalkam, S., Zhou, H., Prabhu, V., Dai, Y., Ryoo, M. S., et al. (2025) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw69036.2025.00644 · Cited by 20
Li, Z., Wu, X., Du, H., Liu, F., Nghiem, H. T., & Shi, G. (2025) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw67362.2025.00147 · Cited by 91
“…[204], MM-Vet [234] Optical Character Recognition (OCR) Test models’ ability to extract objects from visual inputs MM-Vet [234], OCRBench [140], MME [54], MMT-Bench [232] Chart graphic understanding Evaluates models’ ability to interpret graphic-related data infographic VQA [152], AI2D…”
Self-citationDeepEyesV2: Toward Agentic Multimodal ModelPreprint
Hong, J., Zhao, C., Zhu, C., Lu, W., Xu, G., & Yu, X. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.05271 · Cited by 128
Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs
Zhang, Q.-Z., Liu, M.-Z., Li, L.-C., Lu, M., Zhang, Y., Pan, J.-W., She, Q., & Zhang, S.-H. (2025) · Neural Information Processing Systems · doi:10.48550/arxiv.2506.10967 · Cited by 122
“…MMBench [Liu et al., 2025], MMBench-CN [Liu et al., 2025] and MM-Vet [Yu et al., 2023], 4 text-oriented VQA tasks such as TextVQA [Singh et al., 2019], ChartQA [Masry et al., 2022], AI2D [Kembhavi et al., 2016] and OCRBench [Liu et al., 2024e], and 1 multi-turn dialog task MMDU [Liu et al., 2024f].”
X-Omni: Reinforcement Learning Makes Discrete Autoregressive Image Generative Models Great AgainPreprint
Geng, Z., Wang, Y., Ma, Y., Li, C., Rao, Y., Gu, S., Zhong, Z., Lu, Q., Hu, H., Zhang, X., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.22058 · Cited by 116
Self-citationRADIOv2.5: Improved Baselines for Agglomerative Vision Foundation Models
Heinrich, G., Ranzinger, M., Yin, H.-X., Lu, Y., Kautz, J., Tao, A., Catanzaro, B., & Molchanov, P. (2024) · Computer Vision and Pattern Recognition · doi:10.1109/cvpr52734.2025.02094 · Cited by 118
“We leverage insights from Cambrian[42] to select the benchmarks that are most sensitive to the quality of visual features report scores on TextVQA[40], ChartQA[27], DocVQA[29], InfoVQA[28] and OCRBench[26].”
VisuLogic: A Benchmark for Evaluating Visual Reasoning in Multi-modal Large Language ModelsPreprint
Xu, W.-Y., Wang, J., Wang, W.-Y., Chen, Z., Zhou, W.-G., Yang, A.-J., Lu, L., Li, H.-Q., Wang, X.-H., Zhu, X.-Z., et al. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2504.15279 · Cited by 114
“Subsequent works expand the capability coverage of benchmarks into more specialized domains: OCRBench [49], Chartqa [56] and DocVQA [57] assess textual content extraction; AgentBench [48] and ToolEyes [86] test tool usage capabilities; and egocentric perception benchmarks [54, 17] quantify…”
Arif, K. H. I., Yoon, J., Nikolopoulos, D. S., Vandierendonck, H., John, D., & Ji, B. (2024) · AAAI Conference on Artificial Intelligence · doi:10.48550/arxiv.2408.10945 · Cited by 114
“To effectively distribute, we first introduce a hyperparameter α ∈ [0 , 1] to determine the token budget for the full image.”
Dimple: Discrete Diffusion Multimodal Large Language Model with Parallel Decoding
Yu, R., Ma, X., & Wang, X.-C. (2025) · arXiv.org · doi:10.48550/arxiv.2505.16990 · Cited by 110
FastVLM: Efficient Vision Encoding for Vision Language Models
Vasu, P. K. A., Faghri, F., Li, C., Koc, C., True, N., Antony, A., Santhanam, G., Gabriel, J., Grasch, P., Tuzel, O., et al. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.01841 · Cited by 58
“4, we report performance of FastVLM on ChartQA [63], OCRBench [56] and InfoVQA [64] to compare FastVLM against recent methods on text-rich benchmarks.”
Self-citationLuo, G., Yang, X., Dou, W., Wang, Z., Liu, J., Dai, J., Qiao, Y., & Zhu, X. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.02324 · Cited by 23
Scaling Spatial Intelligence with Multimodal Foundation Models
Cai, Z., Wang, R.-S., Gu, C.-Y., Pu, F., Xu, J., Wang, Y.-B., Yin, W.-Q., Yang, Z.-T., Wei, C., Sun, Q., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2511.13719 · Cited by 101
“…training on spatial intelligence data ( i.e. , SenseNova-SI-8M), we evaluate its performance on four additional multimodal benchmarks: MMBench-En [33] and MMStar [9] for holistic multimodal understanding, AI2D [22] for scientific diagram reasoning, and OCRBench [34] for character recognition (OCR).”
InfluentialLu, S.-Y., Li, Y., Xia, Y., Hu, Y.-W., Zhao, S.-S., Ma, Y., Wei, Z.-C., Li, Y.-L., Duan, L.-H., Zhao, J.-S., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2508.11737 · Cited by 101
“It also scores 87.9 on OCRBench, underscoring its precise text recognition, and 65.1 on HallusionBench, indicating fewer hallucinations and higher reliability.”
InfluentialMiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal Interaction
Cui, J.-B., Xu, B.-K., Wang, C., Yu, T., Sun, W.-Y., Xu, Y., Wang, T.-R., He, Z., Ma, W., Cai, T., et al. (2026) · arXiv.org · doi:10.48550/arxiv.2604.27393 · Cited by 100
“On OCRBench [41], TextVQA [42], and DocVQA [43], MiniCPM-o 4.5 is on par with top-tier models.”
InfluentialMonkeyOCR: Document Parsing with a Structure-Recognition-Relation Triplet Paradigm
Li, Z., Liu, Y.-L., Liu, Q., Ma, Z., Zhang, Z., Zhang, S., Yang, B., Guo, Z., Zhang, J.-R., Wang, X., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2506.05218 · Cited by 98
“These findings indicate that while smaller models are sufficient for simpler tasks, a larger parameter capacity is crucial for maintaining performance on complex tasks.”
Self-citationYang, R., Zhu, Z., Li, Y.-W., Huang, J.-J., Yan, S., Zhou, S., Liu, Z., Li, X.-T., Li, S.-Y., Wang, W.-Q., et al. (2025) · European Conference on Computer Vision · doi:10.48550/arxiv.2511.05491 · Cited by 95
InstructVLA: Vision-Language-Action Instruction Tuning from Understanding to ManipulationPreprint
Yang, S., Li, H., Wang, B., Chen, Y., Tian, Y., Wang, T., Wang, H., Zhao, F., Liao, Y., & Pang, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.17520 · Cited by 73
“We use the automatic evaluation from VLMEvalKit (Duan et al., 2024) including MMMU(Val) (Yue et al., 2024), MMStar (Chen et al., 2024a), MME (Fu et al., 2024), OCRBench (Liu et al., 2024e), HallB(Avg) (Guan et al., 2024), MMB(Dev En V1.”
Deng, H., Zou, D., Ma, R., Luo, H.-C., Cao, Y., & Kang, Y. (2025) · arXiv.org · doi:10.48550/arxiv.2503.07065 · Cited by 81
“• OCRBench [27] converts text in images into readable format, is a fundamental task in document understanding.”
InternLM-XComposer2.5-Reward: A Simple Yet Effective Multi-Modal Reward ModelPreprint
Zang, Y., Dong, X., Zhang, P., Cao, Y., Liu, Z., Ding, S., Wu, S., Ma, Y., Duan, H., Zhang, W., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2501.12368 · Cited by 62
“We also report the performance on other categories, such as MMBench [52], MMMU [106] and MMStar [12] for general knowledge, MathVerse [113] and MathVision [89] for math reasoning, TextVQA [79], ChartQA [63] and OCRbench [49] for text-rich document understanding.”
Kwai Keye-VL 1.5 Technical ReportPreprint
Yang, B., Wen, B., Ding, B., Liu, C., Chu, C., Song, C., Rao, C., Yi, C., Li, D., Zang, D., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.01563 · Cited by 94
Self-citationEagle 2.5: Boosting Long-Context Post-Training for Frontier Vision-Language ModelsPreprint
Chen, G., Li, Z., Wang, S., Jiang, J., Liu, Y., Lü, L., Huang, D.-A., Byeon, W., Le, M., Rintamaki, T., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2504.15271 · Cited by 58
“The model also performs well in optical character recognition with 869 on OCRBench [65], comparable to other models in its category.”
Self-citationChain-of-Visual-Thought: Teaching VLMs to See and Think Better with Continuous Visual Tokens
Qin, Y.-M., Wei, B., Ge, J.-X., Kallidromitis, K., Fu, S., Darrell, T., & Wang, X.-D. (2025) · European Conference on Computer Vision · doi:10.48550/arxiv.2511.19418 · Cited by 75
“Besides, we also evaluate C O VT on some non-vision-centric visual benchmarks such as OCRBench [32], MME [17], MUIRBench [48], HallusionBench [21], A-OKVQA [40], TaskMeAny-thing [68], WeMATH [38], and WorldMedQA-V [36].”
VideoChat-R1.5: Visual Test-Time Scaling to Reinforce Multimodal Reasoning by Iterative Perception
Yan, Z., He, Y.-N., Li, X.-H., Yue, Z.-R., Zeng, X.-Y., Wang, Y.-L., Qiao, Y., Wang, L.-M., & Wang, Y. (2025) · Neural Information Processing Systems · doi:10.48550/arxiv.2509.21100 · Cited by 75
“Additionally, on OCRBench [37], the score increases from 856 to 866, reflecting improved fine-grained visual perception.”
Wang, G., Zhao, S.-S., Zhang, X., Cao, L., Zhan, P., Duan, L.-H., Lu, S.-Y., Fu, M.-H., Chen, X., Zhao, J.-S., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2506.23044 · Cited by 75
“…Multi-modal Academic Benchmarks 1 , including MMBench (Liu et al., 2024a), MMStar (Chen et al., 2024b), MMMU-Val (Yue et al., 2024), MathVista-Mini (Lu et al., 2023), HallusionAvg (Guan et al., 2024), AI2D-Test (Kembhavi et al., 2016), OCRBench (Liu et al., 2024b), MMVet (Yu et al., 2024).”
MiMo-VL Technical ReportPreprint
Core Team, Yue, Z., Lin, Z., Song, Y., Wang, W., Ren, S., Gu, S., Li, S., Li, P., Zhao, L., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.03569 · Cited by 173
Self-citationEfficient GPT-4V level multimodal large language model for deployment on edge devices
Yao, Y., Yu, T., Zhang, A., Wang, C., Cui, J., Zhu, H., Cai, T., Chen, C., Li, H., Zhao, W., et al. (2025) · Nature Communications · doi:10.1038/s41467-025-61040-5 · Cited by 60
EMOVA: Empowering Language Models to See, Hear and Speak with Vivid Emotions
Chen, K., Gou, Y., Huang, R., Liu, Z., Tan, D., Xu, J., Wang, C., Zhu, Y., Zeng, Y., Yang, K., et al. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.00513 · Cited by 49
Self-citationLarge multimodal models evaluation: a survey
Zhang, Z., Wang, J., Wen, F., Guo, Y., Zhao, X., Fang, X., Ding, S., Jia, Z., Xiao, J., Shen, Y., et al. (2025) · Science China Information Sciences · doi:10.1007/s11432-025-4676-4 · Cited by 39
Self-citationEgoTextVQA: Towards Egocentric Scene-Text Aware Video Question Answering
Zhou, S., Xiao, J., Li, Q., Li, Y.-C., Yang, X., Guo, D., Wang, M., Chua, T.-S., & Yao, A. (2025) · Computer Vision and Pattern Recognition · doi:10.1109/cvpr52734.2025.00319 · Cited by 64
“Recent advancements in Multimodal Large Language Models (MLLMs) [38, 49, 52] have shown significant potential in addressing scene-text VQA tasks [2, 23, 35].”
EvoCUA: Evolving Computer Use Agents via Learning from Scalable Synthetic Experience
Xue, T., Peng, C., Huang, M.-Q., Guo, L., Han, T., Wang, H., Wang, J., Zhang, X.-C., Yang, X., Zhao, D., et al. (2026) · arXiv.org · doi:10.48550/arxiv.2601.15876 · Cited by 58
Self-citationUniToken: Harmonizing Multimodal Understanding and Generation Through Unified Visual Encoding
Jiao, Y., Qiu, H.-B., Jie, Z.-Q., Chen, S.-X., Chen, J.-J., Ma, L., & Jiang, Y.-G. (2025) · 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw67362.2025.00345 · Cited by 56
MM-IFEngine: Towards Multimodal Instruction Following
Ding, S., Wu, S., Zhao, X., Zang, Y., Duan, H., Dong, X., Zhang, P., Cao, Y., Lin, D., & Wang, J. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.00110 · Cited by 35
“We first take an analysis of the performance of the current open-source MLLMs and proprietary MLLMs on several benchmarks [25, 34], and find that for instruction-following capability, the most powerful open-source MLLM like InternVL2.”
MBQ: Modality-Balanced Quantization for Large Vision-Language Models
Li, S., Hu, Y., Ning, X., Liu, X., Hong, K., Jia, X., Li, X., Yan, Y., Ran, P., Dai, G., et al. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.00394 · Cited by 27
“Specifically, we use OCRBench [32] and TextVQA [40] for text recognition and comprehension, VizWiz [18] and SEED-Bench [19] to test visual perception, and ScienceQA [34] and MMMU [47] to evaluate visual reasoning.”
DeepMMSearch-R1: Empowering Multimodal LLMs in Multimodal Web SearchPreprint
Narayan, K., Xu, Y., Cao, T., Nerella, K., Patel, V. M., Shiee, N., Grasch, P., Jia, C., Yang, Y., & Gan, Z. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.12801 · Cited by 51
“To quantify the impact of SFT+RL training on the proposed model’s general VQA and reasoning ability, we benchmark DeepMMSearchR1-7B (RL) on a range of benchmarks, including OCRBench (Liu et al., 2024d), MMVet (Yu et al., 2023), AI2D (Kembhavi et al., 2016), MathVista (Lu et al., 2024a), MMBench (Liu et al., 2024c), Do…”
Spatial-SSRL: Enhancing Spatial Understanding via Self-Supervised Reinforcement Learning
Liu, Y., Zhang, B., Zang, Y., Cao, Y.-H., Xing, L., Dong, X.-W., Duan, H., Lin, D.-H., & Wang, J. (2025) · arXiv.org · doi:10.48550/arxiv.2510.27606 · Cited by 48
“Second, we evaluate fine-grained visual perception on OCRBench [40], ChartQA test [45], and SeedBench2-plus [31], which require accurate recognition of dense text, charts, and detailed visual information.”
Self-citationKe, W., Zheng, Y., Li, Y., Xu, H., Nie, D., Wang, P., & He, Y. (2025) · ACM Transactions on Information Systems · doi:10.1145/3768156 · Cited by 36
Leveraging large language models for structured information extraction from pathology reports
Balasubramanian, J. B., Adams, D., Roxanis, I., de González, A. B., Coulson, P., Almeida, J. S., & García‐Closas, M. (2025) · Journal of Pathology Informatics · doi:10.1016/j.jpi.2025.100521 · Cited by 31
“35 Evaluating the ability of LLMs tohandle reports with multiple specimen descriptions is another important area for future study.”
Beyond Language Modeling: An Exploration of Multimodal PretrainingPreprint
Tong, S., Fan, D., Nguyen, J., Brown, E., Zhou, G., Qian, S., Zheng, B., Vallaeys, T., Han, J., Fergus, R., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.03276 · Cited by 42
OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations
Ouyang, L., Qu, Y., Zhou, H., Zhu, J., Zhang, R., Lin, Q., Wang, B., Zhao, Z., Jiang, M., Zhao, X., et al. (2025) · 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52734.2025.02313 · Cited by 61
“OCRBench [31] proposes five OCR-related tasks with a greater emphasis on evaluating the model’s visual understanding and reasoning capabilities.”
Self-citationStep-GUI Technical ReportPreprint
GELab-Team, & StepFun (2025) · arXiv.org · doi:10.48550/arxiv.2512.15431 · Cited by 41
Ferret-UI: Grounded Mobile UI Understanding with Multimodal LLMs
You, K., Zhang, H., Schoop, E., Weers, F., Swearngin, A., Nichols, J., Yang, Y., & Gan, Z. (2024) · arXiv · doi:10.48550/arxiv.2404.05719 · Cited by 40
SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify ArchitecturePreprint
Diao, H., Wu, P., Deng, H., Wang, J., Bai, S., Wu, S., Fan, W., Ye, W., Tong, W., Fan, X., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.12500 · Cited by 53
“…(MMMU [159], MMMU-Pro [160], MathVista [87], MathVision [132]), general VLM understanding (MMBench-EN [82], MMStar [16]), OCR (InfoVQA [94], OCRBench [83], OCRBench-v2 [37], AI2D [51]), hallucination detection (HallusionBench [49]), and advanced visual reasoning (BabyVision [15], TiR [68]).”
Xu, M., Gao, M., Li, S., Lu, J., Gan, Z., Lai, Z., Cao, M., Kang, K., Yang, Y., & Dehghan, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.18943 · Cited by 26
“We include OCRBench (Liu et al., 2024d), TextVQA (Singh et al., 2019) and DocVQA (Mathew et al., 2021) measuring OCR, scene text and document understanding, respectively.”
InfluentialChatRex: Taming Multimodal LLM for Joint Perception and Understanding
Jiang, Q., Luo, G., Yang, Y.-Q., Xiong, Y.-D., Zeng, Z.-Y., Chen, Y.-H., Ren, T.-H., & Zhang, L. (2024) · IEEE transactions on multimedia · doi:10.48550/arxiv.2411.18363 · Cited by 40
“…on various academic multimodal benchmarks including MME [23], MMBench [53], SEED I [35], MMstar [11], MM-Vet [96], MMMU [98], AI2D [31], OCRBench [54], TextVQA [77], POPE [40], and HallusionBench [24] shown in Table 3, ChatRex demonstrates strong multimodal capabilities, though there remains a…”
LongCat-Next: Lexicalizing Modalities as Discrete Tokens
Xiao, M., Wang, C., Li, C., Zhang, C., Peng, C., Yu, H., Yang, H., Yan, H., Sun, H., Zhao, H., et al. (2026) · arXiv.org · doi:10.48550/arxiv.2603.27538 · Cited by 39
“To comprehensively evaluate the model’s capability in document understanding and information extraction across diverse document types, we conduct assessments on OCRBench [72], OCRBenchV2 [33], DocVQA [79], and OmniDocBench [88].”
Self-citationAToken: A Unified Tokenizer for VisionPreprint
Lu, J., Song, L., Xu, M., Ahn, B., Wang, Y., Chen, C., Dehghan, A., & Yang, Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.14476 · Cited by 34
“…on 7 standard benchmarks, including RW-QA * , AI2D (Kembhavi et al., 2016), SQA (Lu et al., 2022b), and MMMU (Yue et al., 2024), and MathVISTA (Lu et al., 2024b) for general image QA, as well as OCRBench (Liu et al., 2024a) and TextVQA (Singh et al., 2019) for text and document understanding.”
DocVLM: Make Your VLM an Efficient Reader
Nacson, M. S., Aberdam, A., Ganz, R., Ben-Avraham, E., Golts, A., Kittenplon, Y., Mazor, S., & Litman, R. (2024) · Computer Vision and Pattern Recognition · doi:10.1109/cvpr52734.2025.02701 · Cited by 39
LLaVA-Critic-R1: Your Critic Model is Secretly a Strong Policy ModelPreprint
Wang, X., Li, C., Yang, J., Zhang, K., Liu, B., Xiong, T., & Huang, F. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.00676 · Cited by 36
“We conduct experiment on five multimodel reasoning and understanding benchmarks: MathVista, MathVision, MathVerse, MMMU, and MMStar.”
InfluentialSelf-citationGLM-5V-Turbo: Toward a Native Foundation Model for Multimodal AgentsPreprint
V Team, Hong, W., Gu, X., Pan, Z., Yang, Z., Wang, Y., Wang, Y., Yue, Y., Wang, Y., Wang, Y., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.26752 · Cited by 45
“…[23] and PointBench [6] respectively), video understanding (+5.6% on MVBench [24]), 3D grounding (+7.7% on SUNRGBD [33]), OCR (+4.2% on OCRBench [25]), and chart understanding (+7.7% on CharXiv [40]); on reasoning -heavy tasks such as STEM (+1.8% on MMMU_Val [46], MMMU_Pro [47], MathVista [26]…”
InfluentialTUNA: Taming Unified Visual Representations for Native Unified Multimodal ModelsPreprint
Liu, Z., Ren, W., Liu, H., Zhou, Z., Chen, S., Qiu, H., Huang, X., An, Z., Yang, F., Patel, A., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.02014 · Cited by 36
“…understanding capabilities on nine benchmarks, including general VQA benchmarks such as MME [11], GQA [18], RealWorldQA [66] and SEED-Bench [22]; knowledge-intensive benchmarks such as MMMU [78], MMStar [5], and AI2D [20]; and text-centric benchmarks including ChartQA [36] and OCRBench [32].”
Lei, W., Wang, J., Wang, H., Li, X., Liew, J. H., Feng, J., & Huang, Z. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.01930 · Cited by 26
“Specifically, MLLM benchmarks encompass MMBench-EN [50], SEEDBench-IMG [40], MMVet [85], MME [46], HallusionBench [30], MathVista MINI [53], and OCR-Bench [51].”
MinerU2.5-Pro: Pushing the Limits of Data-Centric Document Parsing at Scale
Wang, B., He, T.-Y., Ouyang, L.-K., Wu, F., Zhao, Z.-Y., Chu, T., Qu, Y., Jin, Z.-J., Zeng, W., Miao, Z.-Y., et al. (2026) · arXiv.org · doi:10.48550/arxiv.2604.04771 · Cited by 35
“OmniDocBench [27] integrates all three metrics and provides one of the most comprehensive document parsing evaluation frameworks to date; OCRBench [20] and CC-OCR [44] focus on evaluating multimodal models’ overall OCR capabilities.”
Self-citationKwai Keye-VL Technical ReportPreprint
Kwai Keye Team, Yang, B., Wen, B., Liu, C., Chu, C., Song, C., Rao, C., Yi, C., Li, D., Zang, D., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.01949 · Cited by 81
Self-citationNemotron 3 Nano Omni: Efficient and Open Multimodal IntelligencePreprint
NVIDIA, :, Deshmukh, A. S., Chumachenko, K., Rintamaki, T., Le, M., Poon, T., Taheri, D. M., Karmanov, I., Liu, G., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.24954 · Cited by 42
“In particular, it attains leading results in document understanding, audio-visual reasoning, and audio benchmarks, ranking at or near the top of leaderboards such as OCRBench-V2 (Liu et al., 2024), MMLongBench-DOC (Ma et al., 2024), VoiceBench (Chen et al., 2024a), WorldSense (Hong et al., 2026), and DailyOmni (Zhou e…”
Self-citationLongCat-Flash-Omni Technical ReportPreprint
Meituan LongCat Team, Wang, B., Bayan, Xiao, B., Zhang, B., Rong, B., Chen, B., Wan, C., Zhang, C., Huang, C. L., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.00279 · Cited by 40
“• Text Recognition & Document Understanding : ChartQA [Masry et al., 2022], DocVQA [Mathew et al., 2021], and OCRBench [Liu et al., 2024c], OmniDocBench EN/ZH [Ouyang et al., 2024].”
InteractiveOmni: A Unified Omni-modal Model for Audio-Visual Multi-turn DialoguePreprint
Tong, W., Guo, H., Ran, D., Chen, J., Lu, J., Wang, K., Li, K., Zhu, X., Li, J., Li, K., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.13747 · Cited by 34
“1 [99], MMStar [24], MMMU [183], MathVista [103], HallusionBench [60], AI2D [79], and OCRBench [100].”
Ming-UniVision: Joint Image Understanding and Generation with a Unified Continuous Tokenizer
Huang, Z.-Y., Zheng, D.-D., Zou, C., Liu, R., Wang, X.-L., Ji, K., Chai, W., Sun, J.-X., Wang, L.-B., Lv, Y.-J., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2510.06590 · Cited by 33
“However, we observe a performance gap on OCRBench (Liu et al., 2024b) and MMMU (Yue et al., 2024), where fine-grained recognition is critical.”
Granite Vision: a lightweight, open-source multimodal model for enterprise IntelligencePreprint
Granite Vision Team, Karlinsky, L., Arbelle, A., Daniels, A., Nassar, A., Alfassy, A., Wu, B., Schwartz, E., Joshi, D., Kondic, J., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.09927 · Cited by 37
MIA-DPO: Multi-Image Augmented Direct Preference Optimization For Large Vision-Language Models
Liu, Z.-Y., Zang, Y., Dong, X.-W., Zhang, P., Cao, Y.-H., Duan, H., He, C.-H., Xiong, Y., Lin, D.-H., & Wang, J. (2024) · International Conference on Learning Representations · doi:10.48550/arxiv.2410.17637 · Cited by 33
“Subse-quently, we also test the model on several single-image benchmarks: MMStar (Chen et al., 2024a), ScienceQA (Lu et al., 2022), MMVet (Yu et al., 2023), POPE (Li et al., 2023c), MMBench (Liu et al., 2023), MathVista (Lu et al., 2023), AI2D (Kembhavi et al., 2016), and OCRBench (Liu et al., 2024c).”
Scaling Language-Centric Omnimodal Representation Learning
Xiao, C.-H., Chan, H.-P., Zhang, H., Xu, W.-W., Aljunied, M., & Rong, Y. (2025) · Neural Information Processing Systems · doi:10.48550/arxiv.2510.11693 · Cited by 32
“Generative performance is measured by averaging results from TextVQA [56], DocVQA [45], OCRBench [42], and ChartQA [44].”
Yang, C., Dong, X., Zhu, X., Su, W., Wang, J., Tian, H., Chen, Z., Wang, W., Lu, L., & Dai, J. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.02322 · Cited by 16
Valley2: Exploring Multimodal Models with Scalable Vision-Language DesignPreprint
Wu, Z., Chen, Z., Luo, R., Zhang, C., Gao, Y., He, Z., Wang, X., Lin, H., & Qiu, M. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2501.05901 · Cited by 25
“Vision and OCR benchmarks (e.g., OCR-Bench [45], ChartQA [58], RealWorldQA) primarily target visual perception and textual recognition.”
Visual-Advantage On-Policy Distillation for Vision-Language Models
Liu, R., Lv, X., Li, G., Zhu, X.-M., Wang, Z., Zhang, Z.-B., Chen, J., Li, Z.-H., Li, B., Gao, J., et al. (2026) · arXiv.org · doi:10.48550/arxiv.2605.21924 · Cited by 30
“We evaluate on eight benchmarks grouped into mathematical reasoning (WeMath [11], MathVista [12], MathVerse [13]) and visual understanding (HallusionBench [14], AI2D [15], MMMU [16], MMStar [17], OCRBench [18]).”
Self-citationSTEP3-VL-10B Technical ReportPreprint
Huang, A., Yao, C., Han, C., Wan, F., Guo, H., Lv, H., Zhou, H., Wang, J., Zhou, J., Sun, J., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.09668 · Cited by 44
“S TEP 3-VL-10B exhibits frontier-class document intelligence, achieving 86.75% on OCRBench and 89.35% on AI2D.”
InfluentialSelf-citationVINO: A Unified Visual Generator with Interleaved OmniModal ContextPreprint
Chen, J., He, T., Fu, Z., Wan, P., Gai, K., & Ye, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.02358 · Cited by 31
Huang, M., Shi, Y., Peng, D., Lai, S., Xie, Z., & Jin, L. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.17163 · Cited by 26
“First, since most of our data collection and annotation processes are performed manually, the high costs associated with these processes have resulted in our dataset size being comparable to previous methods [31, 58].”
InfluentialSelf-citationEfficient Multi-modal Large Language Models via Progressive Consistency Distillation
Wen, Z., Wang, S.-B., Zhou, Y., Zhang, J.-Y., Zhang, Q.-T., Gao, Y., Chen, Z.-R., Wang, B., Li, W.-J., He, C.-H., et al. (2025) · Neural Information Processing Systems · doi:10.48550/arxiv.2510.00515 · Cited by 29
“• OCRBench [46] is a comprehensive benchmark for evaluating the OCR capabilities of multi-modal language models across five key tasks: text recognition, scene text-centric and document-oriented VQA, key information extraction, and handwritten mathematical expression recognition.”
MLLM-CL: Continual Learning for Multimodal Large Language Models
Zhao, H., Zhu, F., Wang, M., Wang, R., Meng, G., & Zhang, Z. (2025) · arXiv.org · doi:10.48550/arxiv.2506.05453 · Cited by 29
“…of the SFT data, we collect the relevant data from LLaVA-OneVision [34], Monkey [37], ScreenQA [22], Screen2Words [58], MultiUI [42], Math-LLaVA [53], MAVIS [71], CLVER [28] and TallyQA [1] and test them on OCRBench [44], MathVista [46], MMTBench-GUI [66] and CV-Bench-Counting [57], respectively.”
Cao, S., Li, J.-Y., Li, X.-H., Pu, Y.-D., Zhu, K., Gao, Y., Luo, S.-Q., Xin, Y., Qin, Q., Zhou, Y., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2512.21675 · Cited by 28
“MMBench [33] evaluates comprehensive perception and reasoning across everyday images, MathVista [35] centers on mathematical and geometric reasoning within visual scenes, and OCRBench [34] specifically tests a model’s capability to recognize and interpret text embedded in images.”
Self-citationBai, S., Cai, Y., Chen, R., Chen, K., Chen, X., Cheng, Z., Deng, L., Ding, W., Gao, C., Ge, C., et al. (2025) · arXiv · doi:10.48550/arxiv.2511.21631 · Cited by 27
“…(Wang et al., 2024b), MathVision-Wild photo (hereafter MathVision WP ) , MathVista (Lu et al., 2023), We-Math (Qiao et al., 2024), MathVerse (Zhang et al., 2024), DynaMath (Zou et al., 2024), Math-VR (Duan et al., 2025), LogicVista (Xiao et al., 2024), VisualPuzzles (Song et al., 2025b),…”
InfluentialSelf-citationDuan, S., Xue, Y.-Q., Wang, W.-H., Sū, Z., Liu, H., Yang, S., Gan, G., Wang, G., Wang, Z., Yan, S., et al. (2026) · arXiv.org · doi:10.48550/arxiv.2603.10910 · Cited by 27
“Besides, GLM-OCR delivers strong performance across text recognition, formula recognition, table parsing, and key information extraction, reaching 94.0 on OCRBench (Text) [17] and 96.5 on UniMERNet [32], and achieving 85.2 on PubTabNet [36] and 86.0 on TEDS 1 .”
InfluentialSelf-citationMME-VideoOCR: Evaluating OCR-Based Capabilities of Multimodal LLMs in Video Scenarios
Shi, Y., Wang, H.-Q., Xie, W.-L., Zhang, H.-Y., Zhao, L.-J., Zhang, Y.-F., Li, X.-F., Fu, C.-Y., Wen, Z.-E., Liu, W.-T., et al. (2025) · Neural Information Processing Systems · doi:10.48550/arxiv.2505.21333 · Cited by 27
“This straightforward yet effective method is widely adopted in previous work [21, 24, 55].”
A Survey on Fine-Grained Multimodal Large Language Models
Peng, Y., Wang, Z., Li, G., Zheng, X., Yin, S., & He, H. (2026) · Chinese Journal of Electronics · doi:10.23919/cje.2025.00.336 · Cited by 10
Ming-Flash-Omni: A Sparse, Unified Architecture for Multimodal Perception and GenerationPreprint
AI, I., :, Ma, B., Zou, C., Du, C., Yan, C., Jin, C., Shen, C., Lian, C., Fan, C., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.24821 · Cited by 36
“Extensive experiments demonstrate that Ming-Flash-Omni achieves comparable performance with leading MLLMs. Vision → Text (Understanding) As shown in”
Robust Multimodal Large Language Models Against Modality Conflict
Zhang, Z., Zhou, W., Zhao, J., & Li, H.-Q. (2025) · International Conference on Machine Learning · doi:10.48550/arxiv.2507.07151 · Cited by 26
“To analyze the alignment tax of our methods, we test the performance changes of the models on a wide range of vision-language tasks after the fine-tuning, including HallusionBench (Guan et al., 2024), MMBench (Liu et al., 2025), MMStar (Chen et al., 2024), MMMU (Yue et al., 2024), MathVista (Lu et al., 2024), OCRBench…”
Lee, J., Park, J., Kim, J., Kim, Y., Oh, J., Oh, J., & Choi, J. (2024) · Annual Meeting of the Association for Computational Linguistics · doi:10.48550/arxiv.2411.09909 · Cited by 26
Quantization-Aware Distillation for NVFP4 Inference Accuracy RecoveryPreprint
Xin, M., Priyadarshi, S., Xin, J., Kartal, B., Vavre, A., Thekkumpate, A. K., Chen, Z., Mahabaleshwarkar, A. S., Shahaf, I., Bercovich, A., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.20088 · Cited by 18
“We believe this is attributed to the simple training pipeline and also the small accuracy drop from PTQ. Kem-bhavi et al., 2016), ChartQA (Masry et al., 2022), DocVQA (Mathew et al., 2021b), InfoVQA (Mathew et al., 2021a), OCRBench (Liu et al., 2024a)”
SCOPE: Saliency-Coverage Oriented Token Pruning for Efficient Multimodel LLMsPreprint
Deng, J., Li, W., Zhou, J. T., & He, Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.24214 · Cited by 10
“To further demonstrate the effectiveness of SCOPE, we conducted additional experiments on more OCR-specific benchmarks including DocVQA [32], ChartQA [31] and OCRBench [28].”
Benchmarking Vision-Language Models on Chinese Ancient Documents: From OCR to Knowledge Reasoning
Yu, H.-Y., Wu, Y.-C., Shi, F., Liao, L., Lu, J., Ge, X., Wang, H., Zhuo, M., Wu, X.-C., Fei, X., et al. (2025) · arXiv.org · doi:10.48550/arxiv.2509.09731 · Cited by 25
“To alleviate the above problem of single tasks, OCR-Bench (Liu et al. 2024b) and its enhanced version OCR-Bench v2 (Fu et al. 2024) have introduced more diverse settings, covering multiple subtasks, including character recognition, text localization, and some document-level question answering and logical reasoning pro…”
Gong, Z., Li, W., Ma, O., Li, S., Wang, Z., Li, S., Ji, J., Yang, X., Luo, G., Yan, J., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.07966 · Cited by 16
“With the expanding capabilities of MLLMs, an increasing number of studies have focused on systematically evaluating their performance across diverse tasks, including caption generation [9, 3, 65], visual question answering (VQA) [22, 44, 17, 71, 36, 37, 48, 38], and code generation [27].”
Yu, W., Yang, Z.-B., Wan, J.-Q., Song, S.-B., Tang, J., Cheng, W., Liu, Y.-L., & Bai, X. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2026.3677075 · Cited by 25
“3) We further investigate the incorporation of SPOT prompting into multimodal large language models (MLLMs) for VsTP, a domain where MLLMs traditionally underper-formed [15], [16], [17].”
Self-citationQianfan-OCR: A Unified End-to-End Model for Document IntelligencePreprint
Dong, D., Zheng, M., Xu, D., Luo, C., Zhuang, B., Li, Y., He, R., Wang, H., Zhang, W., Wang, W., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.13398 · Cited by 4
“Note that these benchmarks include not only pure OCR recognition but also understanding and key information extraction (KIE) tasks, where specialized OCR models generally underperform.”
InfluentialAdaptVision: Efficient Vision-Language Models via Adaptive Visual AcquisitionPreprint
Lin, Z., Liu, Y., Yang, Y., Tao, L., & Ye, D. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.03794 · Cited by 9
“We conduct experiments on several general VQA benchmarks, including ChartQA [25], OCRBench [22], DocVQA [26], MME [7], MMVet [47], RealWorldQA [41], POPE [18], MathVista [24], MathVerse [48].”
Self-citationYu, W., Yang, Z.-B., Liu, Y.-L., & Bai, X. (2025) · IEEE International Conference on Computer Vision · doi:10.1109/iccv51701.2025.00086 · Cited by 23
“To further evaluate DocThinker beyond the Visual CoT Benchmark [16], we assess its performance on OCRBench [12], a widely used benchmark for text-centric multimodal understanding.”
Self-citationAurangzeb, I., & Yoon, J. H. (2025) · Automation in Construction · doi:10.1016/j.autcon.2025.106286 · Cited by 23
Metadata
- Authors
- Liu, Y., Li, Z., Huang, M., Yang, B., Yu, W., Li, C., Yin, X.-C., Liu, C.-L., Jin, L., & Bai, X.
- Affiliations
- Huazhong University of Science and Technology (CN)
- South China University of Technology (CN)
- Microsoft (United States) (US)
- University of Science and Technology Beijing (CN)
- Corresponding author
- Bai, Xiang
- Venue
- Science China Information Sciences
- Published
- 2024-12
- Volume
- 67
- Issue
- 12
- ISSN
- 1869-1919 (ISSN-L)
- 1674-733X
- OpenAlex
- W4405399726
- Cited by
- 167
- References
- 110
- Citations by year
- 2026: 82 · 2025: 64 · 2024: 16
- FWCI
- 22.68
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- Science China Information Sciences (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by-sa)
- arXiv (Cornell University)
- Funders
- National Natural Science Foundation of China (62225603)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Handwritten Text Recognition Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Keywords
- optical character recognition, large multimodal models, evaluation benchmark, key information extraction, handwritten mathematical expression recognition, multilingual text recognition, GPT-4V, Gemini
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
