Pith. sign in

REVIEW 4 major objections 5 minor 3 cited by

Youtu-Parsing: Perception, Structuring and Recognition via High-Parallelism Decoding

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Youtu-Parsing claims that a 2.5B-parameter document parser can decode with a 5–11× speedup via lossless token parallelism while achieving top scores on OmniDocBench v1.5 and olmOCR-bench.

desk verdict A genuinely useful document-parsing model with released weights, but the paper's headline lossless-decoding claim is contradicted by its own ablation table and needs a major fix before it is credible. read the letter →

arxiv 2601.20430 v2 pith:5YX7IDEF submitted 2026-01-28 cs.CV

Haoyu Cao , Kun Yin , Yunfei Wu , Bing Liu , Zhongpeng Cai , Xiaotian Li , Huang Chen , Xin Li
show 11 more authors
Yinsong Liu Deqiang Jiang Xing Sun Yunsheng Wu Qianyu Li Antai Guo Yanzhen Liao Yanqiu Qu Haodong Lin Chengxu He Shuangyin Liu
This is my paper · ORCID
classification cs.CV
keywords documentparsingtokenparallelismquerylosslessspeculativedecodingNaViTvisionencoderHybridMaskedTrainingOmniDocBencholmOCR-bench
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Youtu-Parsing claims that a single, decoupled vision-language model can outperform both modular pipeline tools and far larger generalist VLMs on full-document parsing, while running 5–11× faster than standard autoregressive decoding through a lossless token-parallel scheme. The model splits parsing into shared visual feature extraction, layout analysis, and region-prompted decoding, then accelerates the final decoding with two complementary parallelisms: predicting up to 64 candidate tokens at once and verifying them in a second forward pass, and querying up to five detected regions per pass. If these claims hold, high-throughput document digitization no longer needs to choose between recognition fidelity and inference speed. The paper reports state-of-the-art overall scores of 93.22 on OmniDocBench v1.5 and 80.5±0.9 on olmOCR-bench with a 2.5B-parameter model.

What carries the argument

The load-bearing mechanism is the two-pass token-parallel decoding loop: the model extends the current context with $n=64$ <mask> tokens, produces $n$ candidate tokens in one forward pass, then runs a second forward pass over the candidate sequence and accepts only the tokens up to the first mismatch, guaranteeing bit-level equality with standard autoregressive decoding under greedy selection. The supporting object is the shared NaViT visual feature map, which lets layout analysis and region-prompted decoding reuse the same encoded representation and makes query parallelism natural. A Hybrid Masked Training objective (80% masked samples, 20% unmasked) instills the look-ahead ability while preserving baseline autoregressive integrity.

What would settle it

Run the released model on any page of OmniDocBench v1.5 with token-parallel decoding under greedy selection and compare its output to the same model's sequential greedy output; any page where the two differ would falsify the lossless claim. Alternatively, an independent re-run of the benchmark that does not reproduce the reported overall score of 93.22 would also falsify the central claim.

Watch

Extended reading notes

Core claim

The central discovery is that document parsing output is structured and predictable enough that a model can be trained to guess multiple future tokens at once, and a cheap verification pass can reject any wrong guesses, making the parallel decoding provably identical to sequential greedy decoding. The paper demonstrates this through Hybrid Masked Training, where 80% of fine-tuning samples receive random mask tokens so the model learns multi-token look-ahead, and reports an average acceptance of 10–20 tokens per iteration, translating to a 5–11× speedup (up to ~27× on tables), consistent with the theoretical speedup $S\approx k/2$ where $k$ is the average accepted tokens per iteration. Together with query parallelism, which batches up to five region prompts into one forward pass, the scheme yields end-to-end per-page latency comparable to much smaller specialized models while achieving higher parsing accuracy.

Load-bearing premise

The lossless guarantee of token parallelism holds only when decoding is greedy (argmax) and the two forward passes produce numerically identical logits; any sampling temperature or nondeterminism in attention kernels would break the exact equivalence.

Editorial extensions

If this is right

  • A single 2.5B parameter model reaches 93.22 overall on OmniDocBench v1.5, surpassing both pipeline tools (Marker, MinerU-pipeline) and VLMs up to 241B parameters.
  • Token-parallel decoding yields 5–11× higher token throughput on typical pages, rising to ~27× on tables, with output identical to sequential greedy decoding.
  • Query parallelism adds ~2× end-to-end speed by batching up to five region queries per forward pass, with accuracy stable or slightly improved.
  • The model covers text, formulas, tables, charts, seals, and hierarchical structures, so a single deployment can replace multiple task-specific components.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The verification scheme is a general recipe: any task whose output is greedy-decoded and cheaply re-scored could adopt the same candidate-then-verify loop, so the idea likely transfers beyond documents to code generation or structured data serialization.
  • The claimed bit-level equivalence is contingent on deterministic greedy decoding; if a deployment switches on sampling for diversity, the exact-match verification no longer preserves the target distribution, and the speedup would come with an uncontrolled distribution shift.
  • Hybrid Masked Training could be tested on larger base models; if acceptance rates grow with model scale, the 5–11× speedup would become a lower bound for future systems.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript introduces Youtu-Parsing, a 2.5B-parameter document parsing vision-language model built on a three-stage decoupled architecture: a NaViT-based shared visual feature extractor, a layout analysis stage, and region-prompted decoding with Youtu-LLM-2B. The central technical contribution is a dual-track parallel decoding scheme. Token parallelism appends up to n=64 mask tokens per step, generates candidate tokens in one forward pass, and verifies them with a second pass using the exact-match rule of Eq. (2); the authors claim this is lossless and mathematically equivalent to autoregressive decoding, yielding 5--11x speedups. Query parallelism batches up to five bounding-box queries in one forward pass, claiming an additional ~2x speedup with equivalent output quality. The paper reports state-of-the-art accuracy on OmniDocBench v1.5 (overall 93.22) and olmOCR-bench (80.5 +/- 0.9), plus fine-grained results on text, table, formula, chart, and seal recognition. It also describes hybrid masked training (80/20), a three-stage training recipe with GRPO, an iterative data refinement pipeline, and open releases of code, model weights, and a demo.

Significance. If the central claims held, this would be a practically significant result: a single 2.5B model that outperforms dedicated pipelines and much larger VLMs on two public document-parsing benchmarks, while also delivering a theoretically lossless and empirically fast decoding scheme, is genuinely useful for high-throughput document intelligence. The open release of code, model weights, and a demo is a clear strength, and the use of external benchmarks (OmniDocBench, olmOCR-bench) provides independent grounding for the accuracy claims. However, the lossless-equivalence assertion is not supported by the paper's own ablations: reported scores vary with parallelism degree n and with the attention backend, which is impossible if outputs are bit-identical to autoregressive decoding. Because losslessness is the stated basis for claiming zero accuracy degradation and mathematical equivalence, this inconsistency is load-bearing. The accuracy results on public benchmarks may well survive independently, but the efficiency and equivalence claims need substantial correction or re-evidence.

major comments (4)
  1. [§4.3.1, Table 10] The claimed lossless equivalence is contradicted by the reported OmniDocBench v1.5 scores. Under the exact-match acceptance rule of Eq. (2) with greedy deterministic decoding, the accepted token sequence—and therefore any deterministic benchmark score—cannot depend on the number n of mask tokens or the attention backend. Table 10 shows Flash Attention Overall rising from 92.93 (n=16) to 93.17 (n=32) to 93.30 (n=64), while Eager Attention stays at 92.90 and differs from Flash at every n. Please explain this variation (e.g., sampling, nondeterministic attention kernels, or inconsistent evaluation conditions) and either provide bit-identical outputs across all configurations or revise the 'mathematical equivalence' and 'bit-level equivalence' claims in Sections 2.3.1 and 5.
  2. [§2.3.1, Eq. (2)] The losslessness guarantee is only valid under a deterministic decoding rule. The verification accepts a candidate when it exactly equals the token from the second forward pass; this guarantees identity with standard autoregressive decoding only if tokens are selected by argmax and if the two forward passes are bitwise deterministic. The paper never states the sampling temperature, whether greedy decoding was used in all experiments, or how nondeterminism from FlashAttention is handled. If sampling is used for output diversity, exact-match verification does not preserve the target distribution, and the claimed equivalence in Sections 2.3.1 and 5 fails. Please specify the exact decoding configuration and demonstrate determinism, or narrow the claim.
  3. [§4.3, Tables 10-11, Conclusion] The speedup numbers are internally inconsistent. The abstract and Section 2.3.1 state 5--11x, Table 10 reports a maximum of 11.13x, Table 11 reports up to 26.82x, and the Conclusion states 10--20x. Moreover, Table 11's caption defines the baseline as n=2, but the table includes n=1 rows with 1.00x speedup. Please report one consistent measurement protocol, specify the baseline, and explain how the per-scenario speedup of 26.82x relates to the claimed 5--11x aggregate.
  4. [§4.3.2, Table 12] Query Parallelism is claimed to maintain output quality equivalent to standard decoding with deterministic sequence decomposition and mapping back to layout elements. Yet Table 12 reports the Overall score changing from 88.77 (m=1) to 90.12 (m=5). If the model's token-level outputs were identical across m, a deterministic scorer would give the same Overall score. The variation implies either different evaluation workloads, nondeterministic decoding, or that query batching changes the outputs; in any case the 'quality equivalent' claim needs qualification or correction.
minor comments (5)
  1. [§1, Contributions] In the bullet list of contributions, 'åToken Parallelism' appears to contain a stray character; please correct the typo.
  2. [Table 5 and §4.2.1] Table 5 reports 'Edit Distance Performance' with values around 98-99 for handwritten text, yet the surrounding text calls these 'recognition accuracies.' Please state explicitly whether lower or higher is better and whether the values are 1 - normalized edit distance.
  3. [§4.2.4, Eq. (5)] In the definition of CSS, the symbol d_edit is described only as character-level edit distance; please specify the normalization convention and the handling of empty cells consistently.
  4. [§4.2.1-§4.2.5] The fine-grained evaluation relies on in-house datasets that are not released; please state whether these datasets will be made available to enable independent verification of the tables.
  5. [Table 2] The resolution notation '12288×32×32' in Table 2 is not explained; please clarify the meaning of the three factors.

Circularity Check

0 steps flagged · score 2.0 of 10

External benchmarks and measured speedups ground the central claims; only minor non-load-bearing self-citations, no circular reduction exhibited.

full rationale

Most load-bearing claims are externally grounded and not circular. The OmniDocBench v1.5 and olmOCR-bench results are independent public benchmarks with fixed scoring protocols, so the SOTA claim is not an artifact of the paper's own definitions or fitted parameters. The 5-11x (and up to 26.82x in Table 11) speedups are measured wall-clock latencies, not quantities derived from the acceptance definition; the 'S~k/2' formula is a simple accounting relation between accepted tokens and two forward passes, not a fitted prediction. Self-citations to Youtu-VL [Lab, 2026] and Youtu-LLM-2B [Lab, 2025] describe the base architecture, but the central evaluation is external, so these citations are not load-bearing circularity. The in-house text/table/formula/chart/seal datasets and the proposed CSS metric are evaluation choices; no statement in the paper indicates the model was trained on the same in-house evaluation sets or that the metric is the optimization target, so no fitted-input-called-prediction reduction is exhibited. The lossless-equivalence claim in Section 2.3.1 is guaranteed by construction via Eq. (2)'s exact-match acceptance, which is a protocol invariant rather than an empirical derivation; it would only become circular if used to explain away the n-dependence in Table 10 (Flash Attention overall 92.93 to 93.30), which instead indicates an implementation/evaluation inconsistency (e.g., non-greedy decoding, nondeterministic attention, or inconsistent scoring conditions). That inconsistency is a correctness risk, not a circularity. Accordingly, no significant circularity is found; the minor self-citations do not support any derived result by themselves.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claims rest on a set of free hyperparameters (n=64, m=5, 80/20 mask ratio) and on unstated assumptions about greedy decoding, determinism, and the reliability of upstream model components and benchmark labels. No new physical entities are introduced. The in-house datasets and the CSS/Edge-F1 metrics are authored for this paper, which adds to the circularity burden for the fine-grained performance claims.

free parameters (3)
  • n (number of mask tokens) = 64
    Default token-parallelism degree is set to n=64 in Section 2.3.1. The speedup and acceptance numbers in Tables 10 and 11 depend directly on this hand-chosen value.
  • m (query parallelism batch size) = 5
    Maximum number of bounding boxes processed per forward pass is empirically set to m=5 in Section 2.3.2. The roughly 2x speedup in Table 12 is measured under this assumption.
  • HMT mask ratio (80/20) = 80/20
    Hybrid Masked Training uses 80% masked samples and 20% unmasked samples. This ratio determines how well the model learns look-ahead dependencies while retaining autoregressive performance, and is set by the authors without an ablation.
assumptions (3)
  • domain assumption Greedy/argmax decoding with deterministic forward passes is used at inference.
    Eq. (2) in Section 2.3.1 defines acceptance as exact token equality, which only yields lossless equivalence to standard autoregressive decoding when tokens are selected greedily and when the two forward passes are deterministic. The paper never states its decoding temperature or numerical determinism strategy.
  • domain assumption The pre-trained Youtu-LLM-2B and NaViT/SigLIP vision encoder provide sufficient vision-language priors.
    Section 2.1 builds directly on Youtu-VL and Youtu-LLM-2B. The model's OCR and reasoning capability is largely inherited from these upstream components, whose own performance is taken as given without independent evaluation here.
  • domain assumption The ground-truth labels in OmniDocBench, olmOCR-bench, CC-OCR, OCRBench v2, and the in-house datasets are correct and representative.
    Section 4 treats all benchmark annotations as reliable. The paper does not audit label quality, and for the in-house datasets the annotation process is only described at a high level in Section 3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Youtu-Parsing: Perception, Structuring and Recognition via High-Parallelism Decoding." pith.science (2026). https://pith.science/paper/5YX7IDEF

@misc{pith2026260120430,
  author       = {Pith},
  title        = {Pith review of: Youtu-Parsing: Perception, Structuring and Recognition via High-Parallelism Decoding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5YX7IDEF}},
  note         = {Machine review of arXiv:2601.20430}
}
read the original abstract

This paper presents Youtu-Parsing, an efficient and versatile document parsing model designed for high-performance content extraction. The architecture employs a native Vision Transformer (ViT) featuring a dynamic-resolution visual encoder to extract shared document features, coupled with a prompt-guided Youtu-LLM-2B language model for layout analysis and region-prompted decoding. Leveraging this decoupled and feature-reusable framework, we introduce a high-parallelism decoding strategy comprising two core components: token parallelism and query parallelism. The token parallelism strategy concurrently generates up to 64 candidate tokens per inference step, which are subsequently validated through a verification mechanism. This approach yields a 5--11x speedup over traditional autoregressive decoding and is particularly well-suited for highly structured scenarios, such as table recognition. To further exploit the advantages of region-prompted decoding, the query parallelism strategy enables simultaneous content prediction for multiple bounding boxes (up to five), providing an additional 2x acceleration while maintaining output quality equivalent to standard decoding. Youtu-Parsing encompasses a diverse range of document elements, including text, formulas, tables, charts, seals, and hierarchical structures. Furthermore, the model exhibits strong robustness when handling rare characters, multilingual text, and handwritten content. Extensive evaluations demonstrate that Youtu-Parsing achieves state-of-the-art (SOTA) performance on both the OmniDocBench and olmOCR-bench benchmarks. Overall, Youtu-Parsing demonstrates significant experimental value and practical utility for large-scale document intelligence applications.

Figures

Figures reproduced from arXiv: 2601.20430 by the authors.

Figure 1
Figure 1. Performance of Youtu-Parsing on OmniDocBench v1.5. Youtu-Parsing surpasses both general-purpose vision￾language models and specialized domain models, setting new benchmarks in text recognition, formula recognition, table recognition, and reading order prediction across multiple evaluation tasks [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Parsing Samples 2 [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Youtu-Parsing Framework Overview. The system implements a three-stage cascaded pipeline: (1) Shared Visual Feature Extraction, utilizing a NaViT encoder to extract multi-scale, high-fidelity visual representations; (2) Layout Analysis, which performs cross-modal fusion to precisely identify and categorize structural document elements; (3) Region-prompted Decoding, where localized regions are processed via customized… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The framework of the parallel decoding strategy. Iterative Generation. The accepted tokens are appended to the sequence, and the process iterates. In document parsing, the high predictability of structural patterns—such as LaTeX syntax and table format￾ting—consistentl…
Figure 5
Figure 5. Figure 5: The framework of hierarchical structure analysis. Relational Markup System. We define three specialized markers to characterize the logical associations between document elements: • Parent-Child Relation («): This marker denotes logical subordination. If element B is a…
Figure 6
Figure 6. Figure 6: The iterative data refinement pipeline. Raw documents undergo re-annotation via specialized expert models and large language models (LLMs), followed by consistency-based filtering to ensure label fidelity. High-quality samples are then integrated with synthetic data to…
Figure 7
Figure 7. Figure 7: Overview of the unified synthetic data pipeline for document parsing, including four key stages: (1) Collection of Corpora & Templates, (2) Augmentation of Corpora & Templates, (3) Unified Synthesis Engine, and (4) Generation of Synthetic Multi-Task Data. Each stage co…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. HPD-Parsing: Hierarchical Parallel Document Parsing

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Hierarchical parallel decoding — a global layout branch plus concurrent content branches with multi-token prediction — reaches 4,752 tokens/sec (≈3× a vanilla autoregressive baseline) at competitive accuracy on OmniDocBench.

  2. NaviDC-OCR: Navigating Document Parsing Across Digital and Camera-Captured Documents

    cs.CV 2026-08 conditional novelty 5.0 of 10

    NaviDC-OCR is a 1.2B parameter document parser that combines deformation-aware training, adaptive layout sampling, and content-structure decoupled learning to reach state-of-the-art scores on OmniDocBench v1.6, Wild-O...

  3. OvisOCR2 Technical Report

    cs.CV 2026-07 conditional novelty 4.0 of 10

    A 0.8B end-to-end document-to-Markdown model reportedly tops OmniDocBench v1.6 (96.58) and PureDocBench (75.06), outperforming pipeline-based OCR systems.

Reference graph

Works this paper leans on

59 extracted references · 10 canonical work pages · cited by 3 Pith papers

  1. [1]

    Document parsing unveiled: Techniques, challenges, and prospects for structured information extraction.arXiv preprint arXiv:2410.21169, 2024

    Qintong Zhang, Bin Wang, Victor Shea-Jay Huang, Junyuan Zhang, Zhengren Wang, Hao Liang, Conghui He, and Wentao Zhang. Document parsing unveiled: Techniques, challenges, and prospects for structured information extraction.arXiv preprint arXiv:2410.21169, 2024

    arXiv

  2. [2]

    Dolphin: Document image parsing via heterogeneous anchor prompting.arXiv preprint arXiv:2505.14059, 2025

    Hao Feng, Shu Wei, Xiang Fei, Wei Shi, Yingdong Han, Lei Liao, Jinghui Lu, Binghong Wu, Qi Liu, Chunhui Lin, et al. Dolphin: Document image parsing via heterogeneous anchor prompting.arXiv preprint arXiv:2505.14059, 2025

    arXiv 2025

  3. [3]

    Monkeyocr: Document parsing with a structure-recognition-relation triplet paradigm.arXiv preprint arXiv:2506.05218, 2025

    Zhang Li, Yuliang Liu, Qiang Liu, Zhiyin Ma, Ziyang Zhang, Shuo Zhang, Zidun Guo, Jiarui Zhang, Xinyu Wang, and Xiang Bai. Monkeyocr: Document parsing with a structure-recognition-relation triplet paradigm.arXiv preprint arXiv:2506.05218, 2025

    arXiv 2025

  4. [4]

    Mineru: An open-source solution for precise document content extraction.arXiv preprint arXiv:2409.18839, 2024

    Bin Wang, Chao Xu, Xiaomeng Zhao, Linke Ouyang, Fan Wu, Zhiyuan Zhao, Rui Xu, Kaiwen Liu, Yuan Qu, Fukai Shang, et al. Mineru: An open-source solution for precise document content extraction.arXiv preprint arXiv:2409.18839, 2024

    arXiv 2024

  5. [5]

    olmocr: Unlocking trillions of tokens in pdfs with vision language models.arXiv preprint arXiv:2502.18443, 2025

    Jake Poznanski, Aman Rangapur, Jon Borchardt, Jason Dunkelberger, Regan Huff, Daniel Lin, Christo- pher Wilhelm, Kyle Lo, and Luca Soldaini. olmocr: Unlocking trillions of tokens in pdfs with vision language models.arXiv preprint arXiv:2502.18443, 2025

    arXiv 2025

  6. [6]

    Yumeng Li, Guang Yang, Hao Liu, Bowen Wang, and Colin Zhang. dots. ocr: Multilingual document layout parsing in a single vision-language model.arXiv preprint arXiv:2512.02498, 2025

    arXiv 2025

  7. [7]

    Junbo Niu, Zheng Liu, Zhuangcheng Gu, Bin Wang, Linke Ouyang, Zhiyuan Zhao, Tao Chu, Tianyao He, Fan Wu, Qintong Zhang, et al. Mineru2. 5: A decoupled vision-language model for efficient high-resolution document parsing.arXiv preprint arXiv:2509.22186, 2025

    arXiv 2025

  8. [8]

    Paddleocr-vl: Boosting multilingual document parsing via a 0.9 b ultra-compact vision-language model.arXiv preprint arXiv:2510.14528, 2025

    Cheng Cui, Ting Sun, Suyin Liang, Tingquan Gao, Zelun Zhang, Jiaxuan Liu, Xueqing Wang, Changda Zhou, Hongen Liu, Manhui Lin, et al. Paddleocr-vl: Boosting multilingual document parsing via a 0.9 b ultra-compact vision-language model.arXiv preprint arXiv:2510.14528, 2025

    arXiv

Show all 59 references
  1. [9]

    Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901, 2020

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901, 2020

    1901

  2. [10]

    Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971, 2023

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971, 2023. †Corresponding aut...

    2023 arXiv

  3. [11]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  4. [12]

    Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    2023 arXiv

  5. [13]

    Mistral 7b.arXiv preprint arXiv:2310.06825, 2023

    Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, et al. Mistral 7b.arXiv preprint arXiv:2310.06825, 2023

    2023 arXiv

  6. [14]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational conference on machine learning, pag...

    2021

  7. [15]

    Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    2023

  8. [16]

    Minigpt-4: Enhancing vision-language understanding with advanced large language models.arXiv preprint arXiv:2304.10592, 2023

    Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mohamed Elhoseiny. Minigpt-4: Enhancing vision-language understanding with advanced large language models.arXiv preprint arXiv:2304.10592, 2023

    2023 arXiv

  9. [17]

    Instructblip: Towards general-purpose vision-language models with instruction tuning.Advances in neural information processing systems, 36:49250–49267, 2023

    Wenliang Dai, Junnan Li, Dongxu Li, Anthony Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale N Fung, and Steven Hoi. Instructblip: Towards general-purpose vision-language models with instruction tuning.Advances in neural information processing systems, 36:49250–49267, 2023

    2023

  10. [18]

    Siglip 2: Multilingual vision-language encoders with improved semantic understanding, localization, and dense features

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, et al. Siglip 2: Multilingual vision-language encoders with improved semantic understanding, localization, and ...

    2025 arXiv

  11. [19]

    Paddleocr 3.0 technical report.arXiv preprint arXiv:2507.05595, 2025

    Cheng Cui, Ting Sun, Manhui Lin, Tingquan Gao, Yubo Zhang, Jiaxuan Liu, Xueqing Wang, Zelun Zhang, Changda Zhou, Hongen Liu, et al. Paddleocr 3.0 technical report.arXiv preprint arXiv:2507.05595, 2025

    2025 arXiv

  12. [20]

    Docling: An efficient open- source toolkit for ai-driven document conversion.arXiv preprint arXiv:2501.17887, 2025

    Nikolaos Livathinos, Christoph Auer, Maksym Lysak, Ahmed Nassar, Michele Dolfi, Panos Vagenas, Cesar Berrospi Ramis, Matteo Omenetti, Kasper Dinkla, Yusik Kim, et al. Docling: An efficient open- source toolkit for ai-driven document conversion.arXiv preprint arXiv:2501.17887, 2025

    2025 arXiv

  13. [21]

    Deepseek-ocr: Contexts optical compression.arXiv preprint arXiv:2510.18234, 2025

    Haoran Wei, Yaofeng Sun, and Yukun Li. Deepseek-ocr: Contexts optical compression.arXiv preprint arXiv:2510.18234, 2025

    2025 arXiv

  14. [22]

    Learning to recover from multi-modality errors for non-autoregressive neural machine translation.arXiv preprint arXiv:2006.05165, 2020

    Qiu Ran, Yankai Lin, Peng Li, and Jie Zhou. Learning to recover from multi-modality errors for non-autoregressive neural machine translation.arXiv preprint arXiv:2006.05165, 2020

    2006 arXiv

  15. [23]

    Maskgit: Masked generative image transformer

    Huiwen Chang, Han Zhang, Lu Jiang, Ce Liu, and William T Freeman. Maskgit: Masked generative image transformer. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11315–11325, 2022

    2022

  16. [24]

    Medusa: Simple llm inference acceleration framework with multiple decoding heads.arXiv preprint arXiv:2401.10774, 2024

    Tianle Cai, Yuhong Li, Zhengyang Geng, Hongwu Peng, Jason D Lee, Deming Chen, and Tri Dao. Medusa: Simple llm inference acceleration framework with multiple decoding heads.arXiv preprint arXiv:2401.10774, 2024

    2024 arXiv

  17. [25]

    Youtu-vl: Unleashing visual potential via unified vision-language supervision

    Tencent Youtu Lab. Youtu-vl: Unleashing visual potential via unified vision-language supervision. 2026. URLhttps://arxiv.org/abs/2601.19798

    2026

  18. [26]

    Youtu-llm: Unlocking the native agentic potential for lightweight large language models.arXiv preprint arXiv:2512.24618, 2025

    Tencent Youtu Lab. Youtu-llm: Unlocking the native agentic potential for lightweight large language models.arXiv preprint arXiv:2512.24618, 2025. URLhttps://arxiv.org/abs/2512.24618. 24 Youtu-Parsing Technical Report

    2025

  19. [27]

    Optimized table tokenization for table structure recognition

    Maksym Lysak, Ahmed Nassar, Nikolaos Livathinos, Christoph Auer, and Peter Staar. Optimized table tokenization for table structure recognition. InInternational Conference on Document Analysis and Recognition, pages 37–50. Springer, 2023

    2023

  20. [28]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

    2024 arXiv

  21. [29]

    Omnidocbench: Benchmarking diverse pdf document parsing with comprehensive annotations

    Linke Ouyang, Yuan Qu, Hongbin Zhou, Jiawei Zhu, Rui Zhang, Qunshu Lin, Bin Wang, Zhiyuan Zhao, Man Jiang, Xiaomeng Zhao, et al. Omnidocbench: Benchmarking diverse pdf document parsing with comprehensive annotations. InProceedings of the Computer Vision and Pattern Recognition...

    2025

  22. [30]

    Marker.https://github.com/datalab-to/marker, 2025

    Vik Paruchuri. Marker.https://github.com/datalab-to/marker, 2025

    2025

  23. [31]

    Gpt-4o system card.arXiv preprint arXiv:2410.21276, 2024

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv preprint arXiv:2410.21276, 2024

    2024 arXiv

  24. [32]

    Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models.arXiv preprint arXiv:2504.10479, 2025

    Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shenglong Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Jie Shao, et al. Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models.arXiv preprint arXiv:2504.10479, 2025

    2025 arXiv

  25. [34]

    Qwen2.5-vl technical report.arXiv preprint arXiv:2502.13923, 2025

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Han...

    2025 arXiv

  26. [35]

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261, 2025

    Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, et al. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabil...

    2025 arXiv

  27. [36]

    Ocrflux.https://github.com/chatdoc- com/OCRFlux, 2025

    chatdoc com. Ocrflux.https://github.com/chatdoc- com/OCRFlux, 2025

    2025

  28. [37]

    Mistral-ocr.https://mistral.ai/news/mistral-ocr?utm source=ai-bot.cn, 2025

    Mistral AI Team. Mistral-ocr.https://mistral.ai/news/mistral-ocr?utm source=ai-bot.cn, 2025

    2025

  29. [38]

    Points-reader: Distillation-free adaptation of vision-language models for document conversion

    Yuan Liu, Zhongyin Zhao, Le Tian, Haicheng Wang, Xubing Ye, Yangxiu You, Zilin Yu, Chuhan Wu, Zhou Xiao, Yang Yu, et al. Points-reader: Distillation-free adaptation of vision-language models for document conversion. InProceedings of the 2025 Conference on Empirical Methods in ...

    2025

  30. [39]

    Nanonets-ocr-s: A model for transforming documents into structured markdown with intelligent content recognition and semantic tagging, 2025

    Souvik Mandal, Ashish Talewar, Paras Ahuja, and Prathamesh Juvatkar. Nanonets-ocr-s: A model for transforming documents into structured markdown with intelligent content recognition and semantic tagging, 2025

    2025

  31. [40]

    Image over text: Transforming formula recognition evaluation with character detection matching

    Bin Wang, Fan Wu, Linke Ouyang, Zhuangcheng Gu, Rui Zhang, Renqiu Xia, Botian Shi, Bo Zhang, and Conghui He. Image over text: Transforming formula recognition evaluation with character detection matching. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern ...

    2025

  32. [41]

    General ocr theory: Towards ocr-2.0 via a unified end-to-end model

    Haoran Wei, Chenglong Liu, Jinyue Chen, Jia Wang, Lingyu Kong, Yanming Xu, Zheng Ge, Liang Zhao, Jianjian Sun, Yuang Peng, et al. General ocr theory: Towards ocr-2.0 via a unified end-to-end model. arXiv preprint arXiv:2409.01704, 2024. 25 Youtu-Parsing Technical Report

    2024 arXiv

  33. [42]

    Google deepmind

    Gemini Team. Google deepmind. gemini 2.5: Pushing the frontier with advanced reasoning, multi- modality, long context, and next generation agentic capabilities. Technical report, Technical Report v2. 5, Google DeepMind, 2025

    2025

  34. [43]

    Cc-ocr: A comprehensive and challenging ocr benchmark for evaluating large multimodal models in literacy

    Zhibo Yang, Jun Tang, Zhaohai Li, Pengfei Wang, Jianqiang Wan, Humen Zhong, Xuejing Liu, Mingkun Yang, Peng Wang, Shuai Bai, et al. Cc-ocr: A comprehensive and challenging ocr benchmark for evaluating large multimodal models in literacy. InProceedings of the IEEE/CVF Internati...

    2025

  35. [44]

    Ocrbench v2: An improved benchmark for evaluating large multimodal models on visual text localization and reasoning.arXiv preprint arXiv:2501.00321, 2024

    Ling Fu, Zhebin Kuang, Jiajun Song, Mingxin Huang, Biao Yang, Yuzhe Li, Linghao Zhu, Qidi Luo, Xinyu Wang, Hao Lu, et al. Ocrbench v2: An improved benchmark for evaluating large multimodal models on visual text localization and reasoning.arXiv preprint arXiv:2501.00321, 2024

    2024 arXiv

  36. [45]

    Qwen3 technical report, 2025

    Qwen Team. Qwen3 technical report, 2025. URLhttps://arxiv.org/abs/2505.09388

    2025 arXiv

  37. [46]

    Onechart: Purify the chart structural extraction via one auxiliary token

    Jinyue Chen, Lingyu Kong, Haoran Wei, Chenglong Liu, Zheng Ge, Liang Zhao, Jianjian Sun, Chunrui Han, and Xiangyu Zhang. Onechart: Purify the chart structural extraction via one auxiliary token. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 147–155, 2024

    2024

  38. [47]

    Deplot: One-shot visual language reasoning by plot-to-table translation

    Fangyu Liu, Julian Eisenschlos, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Wenhu Chen, Nigel Collier, and Yasemin Altun. Deplot: One-shot visual language reasoning by plot-to-table translation. InFindings of the Association for Computational Li...

    2023

  39. [48]

    Doclayout-yolo: Enhancing document layout analysis through diverse synthetic data and global-to-local adaptive perception.arXiv preprint arXiv:2410.12628, 2024

    Zhiyuan Zhao, Hengrui Kang, Bin Wang, and Conghui He. Doclayout-yolo: Enhancing document layout analysis through diverse synthetic data and global-to-local adaptive perception.arXiv preprint arXiv:2410.12628, 2024

    2024 arXiv

  40. [49]

    Pp-doclayout: A unified document layout detection model to accelerate large-scale data construction.arXiv preprint arXiv:2503.17213, 2025

    Ting Sun, Cheng Cui, Yuning Du, and Yi Liu. Pp-doclayout: A unified document layout detection model to accelerate large-scale data construction.arXiv preprint arXiv:2503.17213, 2025

    2025 arXiv

  41. [50]

    An end-to-end formula recognition method integrated attention mechanism.Mathematics, 11(1):177, 2022

    Mingle Zhou, Ming Cai, Gang Li, and Min Li. An end-to-end formula recognition method integrated attention mechanism.Mathematics, 11(1):177, 2022

    2022

  42. [51]

    Deeptabstr: Deep learning based table structure recognition

    Shoaib Ahmed Siddiqui, Imran Ali Fateh, Syed Tahseen Raza Rizvi, Andreas Dengel, and Sheraz Ahmed. Deeptabstr: Deep learning based table structure recognition. In2019 international conference on document analysis and recognition (ICDAR), pages 1403–1409. IEEE, 2019

    2019

  43. [52]

    Tsrformer: Table structure recognition with transformers

    Weihong Lin, Zheng Sun, Chixiang Ma, Mingze Li, Jiawei Wang, Lei Sun, and Qiang Huo. Tsrformer: Table structure recognition with transformers. InProceedings of the 30th ACM International Conference on Multimedia, pages 6473–6482, 2022

    2022

  44. [53]

    Layoutreader: Pre-training of text and layout for reading order detection.arXiv preprint arXiv:2108.11591, 2021

    Zilong Wang, Yiheng Xu, Lei Cui, Jingbo Shang, and Furu Wei. Layoutreader: Pre-training of text and layout for reading order detection.arXiv preprint arXiv:2108.11591, 2021

    2021 arXiv

  45. [54]

    Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks

    Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, et al. Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks. InProceedings of the IEEE/CVF Conference on Computer Visi...

    2024

  46. [55]

    How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites.arXiv preprint arXiv:2404.16821, 2024

    Zhe Chen, Weiyun Wang, Hao Tian, Shenglong Ye, Zhangwei Gao, Erfei Cui, Wenwen Tong, Kongzhi Hu, Jiapeng Luo, Zheng Ma, et al. How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites.arXiv preprint arXiv:2404.16821, 2024

    2024 arXiv

  47. [56]

    Enhancing the reasoning ability of multimodal large language models via mixed preference optimization.arXiv preprint arXiv:2411.10442, 2024

    Weiyun Wang, Zhe Chen, Wenhai Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Jinguo Zhu, Xizhou Zhu, Lewei Lu, Yu Qiao, and Jifeng Dai. Enhancing the reasoning ability of multimodal large language models via mixed preference optimization.arXiv preprint arXiv:2411.10442, 2024. 26 Y...

    2024 arXiv

  48. [57]

    Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling.arXiv preprint arXiv:2412.05271, 2024

    Zhe Chen, Weiyun Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Erfei Cui, Jinguo Zhu, Shenglong Ye, Hao Tian, Zhaoyang Liu, et al. Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling.arXiv preprint arXiv:2412.05271, 2024

    2024 arXiv

  49. [58]

    Internvl3.5: Advancing open-source multimodal models in versatility, reasoning, and efficiency.arXiv preprint arXiv:2508.18265, 2025

    Weiyun Wang, Zhangwei Gao, Lixin Gu, Hengjun Pu, Long Cui, Xingguang Wei, Zhaoyang Liu, Linglin Jing, Shenglong Ye, Jie Shao, et al. Internvl3.5: Advancing open-source multimodal models in versatility, reasoning, and efficiency.arXiv preprint arXiv:2508.18265, 2025

    2025 arXiv

  50. [59]

    Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond.arXiv preprint arXiv:2308.12966, 2023

    Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond.arXiv preprint arXiv:2308.12966, 2023

    2023 arXiv

  51. [60]

    Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution.arXiv preprint arXiv:2409.12191, 2024

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. Qwen2-vl: Enhancing vision-language model’s pe...

    2024 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.