Pith. sign in

REVIEW 3 major objections 52 references

Interleaved text-image reasoning works best when both modalities are trained together as one decision process under a single RL objective.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 00:13 UTC pith:FGGCDSPG

load-bearing objection Solid joint-RL recipe for multi-turn text-image CoT; the VLM-judge credit signal is useful but unvalidated, not fatal. the 3 major comments →

arxiv 2607.03748 v1 pith:FGGCDSPG submitted 2026-07-04 cs.AI

Bridging Interleaved Multi-Modal Reasoning as a Unified Decision Process

Zican Hu , Xuyang Hu , Yiming Liu , Zuwei Long , Wei Liu , Yunzhuo Hao , Jiawei Gu , Linjie Li
show 5 more authors
Yu Cheng Zhenhong Sun Weibo Gu Xing Sun Zhi Wang
This is my paper
classification cs.AI
keywords unified multi-modal modelsinterleaved text-image reasoningreinforcement learningMarkov decision processflow matchingvision-thinking process rewardspatial reasoningpolicy gradient
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Unified multi-modal models can alternate verbal thinking with intermediate image generation, but most RL methods only train the text steps and leave image generation outside the loop. This paper claims that treating the full multi-turn text-image-text trajectory as one Markov decision process lets a shared reward signal improve both branches at once. A common trajectory-level advantage is fed into text via ordinary token policy gradients and into images via a flow-matching style update, while a vision-language judge scores each intermediate picture for how much it actually helps the final answer. On spatial reasoning and visual perception tests, the resulting 7B model gains several points over its base and beats a strong proprietary system on several hard spatial tasks. The practical point is that half-optimized RL leaves the visual half of interleaved reasoning undertrained, and closing that loop measurably raises the ceiling of what the model can solve.

Core claim

The paper establishes that casting multi-turn text-image-text chain-of-thought as a single two-level MDP, computing one shared trajectory advantage, and routing it through modality-native gradients (clipped policy gradient on text tokens and reward-weighted flow-matching on image denoising) plus dense turn-level scores from a VLM judge is necessary for effective interleaved multi-modal reasoning; without joint image-branch optimization the gains collapse, and the method lifts a 7B hybrid model by +5.73 average points across seven benchmarks, with large jumps on SAT and V*Bench.

What carries the argument

BRAID: a unified MDP over interleaved turns that shares one trajectory-level advantage and back-propagates it separately through GRPO-style token updates for text and DiffusionNFT-style flow-matching losses for images, augmented by a four-criterion vision-thinking process reward from an external VLM judge.

Load-bearing premise

An external vision-language model scoring intermediate images on four fixed criteria will reliably mark which visual steps actually help the final answer rather than reward superficial or misleading pictures.

What would settle it

Train the identical joint-MDP setup with the image branch frozen or with the VLM process reward replaced by pure terminal reward only, and check whether the large gains on SAT and V*Bench disappear and whether majority-vote accuracy still scales with more samples; if joint image optimization and the dense judge scores add nothing under that controlled ablation, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Policy gradients can cross the text-image boundary end-to-end, so intermediate images become trainable reasoning steps rather than frozen decorations.
  • Dense turn-level visual scores reduce long-horizon credit assignment failure on multi-step spatial and perception tasks.
  • A 7B hybrid model can surpass larger understanding-only systems on viewpoint-heavy and fine-grained visual benchmarks once both modalities are jointly optimized.
  • Improvements in the image branch transfer to standalone text-to-image generation, improving spatial composition and count fidelity.
  • Half-optimized RL that freezes image generation is strictly weaker than full-trajectory optimization on the same backbone.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the external judge is the main source of dense signal, replacing it with a cheap on-policy process reward model would remove a costly dependency and make the method more reproducible.
  • The same shared-advantage pattern should apply to fully autoregressive unified models that emit discrete image tokens, not only hybrid AR-diffusion backbones.
  • Adaptive rather than fixed interleaving patterns could further raise sample efficiency once credit can already flow through both modalities.
  • The approach implies that any multi-modal CoT system whose visual steps are only supervised is leaving a large fraction of its RL potential unused.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 0 minor

Summary. The paper proposes BRAID, a framework that casts multi-turn interleaved text-image-text (TIT-CoT) reasoning in unified multi-modal models as a two-level Markov decision process, so that text and image turns can be optimized jointly under one RL objective. A shared trajectory-level advantage is propagated through modality-native surrogates—GRPO-style clipped policy gradients for autoregressive text tokens and a DiffusionNFT-style reward-aware flow-matching loss for image denoising—while a VLM judge supplies a four-dimensional vision-thinking process reward to densify credit assignment on intermediate images. Built on BAGEL-7B and trained with an SFT-then-RL pipeline, BRAID reports +5.73 average points over BAGEL across seven spatial-reasoning and visual-perception benchmarks, with large gains on SAT and V*Bench, ablations isolating LNFT and r_vis, Maj@n scaling curves, data-allocation and rollout-ratio studies, and qualitative transfer to standalone text-to-image generation.

Significance. If the results hold, this is a meaningful step for post-training of hybrid AR-diffusion UMMs: prior work either freezes image generation during RL (COOPER, DeepEyes) or jointly optimizes only single-turn text-to-image (UniGRPO). The two-level MDP factorization and the explicit channeling of one advantage into modality-native gradients are clean and reusable design choices. The experimental package is stronger than average for this area—component ablations, sampling-budget scaling, data-regime and batch-ratio analyses, and case studies that separate correct answers from faithful visual evidence. Limitations are stated honestly (hybrid backbone only; external VLM judge; fixed interleaving). The main significance risk is that part of the claimed “vision-thinking guidance” may be opaque distillation from a proprietary judge rather than validated credit assignment; even so, the joint image-branch optimization result would remain of independent interest.

major comments (3)
  1. Section 4.4 and Eqs. (9)–(10): the vision-thinking process reward is load-bearing for the abstract’s and conclusion’s claim that “vision-thinking guidance is essential,” yet no validation is reported that the four GPT-5.2 scores (especially r^(ru) and r^(tr)) correlate with terminal correctness, human judgments of intermediate utility, or even inter-run consistency of the judge. The ablation drop when r_vis is removed (Table 4: 65.19 → 62.68) shows the signal is used, not that it is the right signal. Please add at least one of: (i) correlation of per-turn r_vis components with final answer correctness within groups, (ii) human agreement on a scored subset, or (iii) a controlled swap/noise study of the judge scores. Without this, the process-reward half of the central narrative remains an untested axiom rather than a demonstrated credit-assignment mechanism.
  2. Table 1 and Section 5.1: main results and the +5.73 avg. claim are reported from single runs with no error bars, confidence intervals, or multi-seed statistics. Several absolute margins (e.g., CV-Bench 3D −1.24, BLINK +3.00) are small enough that seed variance could reorder methods. Given that the paper’s empirical case for “consistently outperforms” rests on these numbers, please report mean±std over at least 3 seeds for BRAID, SFT, and the two ablations on the seven-benchmark average (and ideally on SAT and V*Bench).
  3. Section 4.3, Eqs. (7)–(8): the mapping from shared advantage  to the soft NFT label r_k = σ(Â/υ) and the scalar balances λ, λ_vis are introduced without sensitivity analysis. Table 3 lists many free parameters (β=0.4, clip ratios, CFG scales, equal 0.25 component weights, etc.), yet only rollout-to-update ratio and data regime are ablated. At minimum, report sensitivity of the seven-benchmark average to λ_vis and to the temperature υ (or to equal vs. learned component weights in Eq. (9)), so that the joint objective is shown to be robust rather than tuned to a single operating point.

Circularity Check

0 steps flagged

No circularity: empirical RL method with external rewards and held-out benchmarks; no derivation reduces to its inputs by construction.

full rationale

BRAID is an empirical systems paper that formulates interleaved TIT-CoT as a two-level MDP (Sec. 4.2), channels a shared trajectory-level advantage into modality-native surrogates (GRPO for text tokens, DiffusionNFT for flow-matching image turns; Eqs. 7–8), and adds a dense process reward from an external proprietary VLM judge (GPT-5.2 scoring four hand-specified criteria; Sec. 4.4, Eqs. 9–10, App. A.4). Terminal rewards are rule-based verifiers on final answers; process rewards are independent of the parameters being optimized. Performance claims are measured on external spatial/perception benchmarks under VLMEvalKit, with ablations isolating LNFT and r_vis. Self-citations (e.g., ThinkMorph for data/SFT recipes, BAGEL as backbone) supply architecture and data, not uniqueness theorems or fitted constants that force the reported gains. No equation equates a claimed prediction to a fitted input; no ansatz is smuggled via self-citation as a uniqueness result. The derivation chain is therefore self-contained and non-circular. (Unvalidated judge quality is a correctness/validity concern, not circularity.)

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 2 invented entities

The central empirical claim rests on standard RL and generative-model machinery plus a handful of hand-chosen scalars and the assumption that an external VLM can score intermediate images usefully. No new physical entities are postulated; the free parameters are ordinary training hyper-parameters and reward weights.

free parameters (5)
  • NFT mixing coefficient β = 0.4
    Controls how strongly the current policy is mixed with the behavior policy inside DiffusionNFT; set to 0.4 without sensitivity analysis.
  • modality balance λ and process-reward weight λ_vis
    Scalar multipliers that trade off text vs image losses and terminal vs process advantages; chosen by authors.
  • soft-label temperature υ
    Maps trajectory advantage into the [0,1] reward label for DiffusionNFT; free temperature.
  • four process-reward component weights = 0.25 each
    Equal 0.25 weights on visual correctness, fidelity, utility, trustworthiness; equal weighting is an untested design choice.
  • rollout-to-update ratio and group size G = 2:1, G=8
    Default 2:1 (64/32) and G=8; ablation shows sensitivity, so the reported gains depend on these choices.
axioms (4)
  • domain assumption Flow-matching denoising can be treated as an MDP whose micro-actions are velocity predictions, allowing DiffusionNFT-style reward injection without likelihoods.
    Taken from prior image-generation RL literature (DiffusionNFT, FlowGRPO) and used without re-derivation in §3–4.3.
  • ad hoc to paper A proprietary VLM (GPT-5.2) can produce integer scores in [1,10] on four criteria that are sufficiently calibrated to serve as dense process rewards.
    Core of the vision-thinking reward (§4.4); no human correlation study or alternative-judge ablation is provided.
  • domain assumption Group-normalized advantages (GRPO-style) remain valid when the same scalar is broadcast to both discrete token actions and continuous denoising paths.
    Assumed by the unified objective in Eq. (7); standard in language RL but unproven for hybrid continuous-discrete trajectories.
  • domain assumption The hybrid AR-diffusion backbone of BAGEL permits independent modality-native gradients while sharing a multi-modal attention backbone.
    Architectural premise of the entire method; experiments are confined to this backbone.
invented entities (2)
  • vision-thinking process reward (four-dimensional VLM score vector) no independent evidence
    purpose: Supplies dense turn-level credit assignment for intermediate images so that long-horizon trajectories do not rely solely on sparse terminal reward.
    Newly defined multi-criteria scoring protocol; independent evidence is limited to the ablation showing performance drop when it is removed.
  • two-level (turn/step) MDP for TIT-CoT no independent evidence
    purpose: Homogenizes text chunks and images as macro-actions so a single trajectory advantage can be defined and then expanded into modality-native micro-actions.
    Formalization introduced in §4.2; it is a modeling choice rather than a discovered physical entity.

pith-pipeline@v1.1.0-grok45 · 26509 in / 3348 out tokens · 28037 ms · 2026-07-12T00:13:16.693241+00:00 · methodology

0 comments
read the original abstract

Unified multi-modal models (UMMs) have shown promising interleaved text-image reasoning capabilities, yet effectively optimizing such multi-turn generation via reinforcement learning (RL) remains an open challenge. Existing approaches apply RL exclusively to text steps, relegating image generation to supervised surrogates, preventing policy gradients from propagating through the full interleaved trajectory across heterogeneous modalities. This leaves the potential of RL for UMMs largely untapped. In the paper, we introduce \textbf{BRAID} (\textbf{B}ridging inte\textbf{R}le\textbf{A}ved mult\textbf{I}-modal reasoning as a unified \textbf{D}ecision process), a simple framework that casts multi-turn text-image-text reasoning as a unified Markov decision process (MDP), enabling joint optimization of textual and visual generation via a single, principled RL objective. BRAID computes a shared trajectory-level advantage and propagates it coherently into both text tokens and image denoising paths, each optimized through its modality-native policy gradient mechanism. To further address long-horizon credit assignment, BRAID employs a vision-language model (VLM) judge that scores each intermediate image on its reasoning utility, supplying dense turn-level feedback to sharpen learning at critical visual branches. Experiments on spatial reasoning and visual perception benchmarks show that BRAID consistently outperforms various baselines, confirming that a unified MDP formulation with vision-thinking guidance is essential for effective multi-modal reasoning.

Figures

Figures reproduced from arXiv: 2607.03748 by Jiawei Gu, Linjie Li, Weibo Gu, Wei Liu, Xing Sun, Xuyang Hu, Yiming Liu, Yu Cheng, Yunzhuo Hao, Zhenhong Sun, Zhi Wang, Zican Hu, Zuwei Long.

Figure 1
Figure 1. Figure 1: SFT, Half-optimized RL, and BRAID. Recent emergence of unified multi-modal models (UMMs) [1, 2] that seamlessly bridge understand￾ing and generation across modalities within a single framework has opened a new frontier in multi-modal reasoning: interleaved Text-Image￾Text Chain-of-Though (TIT-CoT), where verbal thinking alternates with intermediate image generation to externalize perceptual steps beyond th… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of BRAID. (a) Interleaved TIT-CoT reasoning formulated as a two-level MDP. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Maj@n and Avg@n curves on VStar and SAT benchmarks with varying number of samples n = {3, 5, 9, 17}. Complete results are available in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Ablation results. Left: Final test accuracy across seven benchmarks. Middle: Training [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Complete results are available in Table 7. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Complete results are available in [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Text-to-image generation comparison. Interleaved Reasoning. We qualitatively compare BRAID and its ablations on a visual search task (Appendix C). BRAID generates diverse, high-fidelity intermediate images that directly support correct reasoning. In contrast, w/o LNFT produces distorted visuals that induce hallucinated text reasoning, and w/o r vis intro￾duces fabricated artifacts. Although all variants re… view at source ↗
Figure 8
Figure 8. Figure 8: Text-to-image generation comparison across prompts testing spatial compositionality (Row [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

52 extracted references · 27 linked inside Pith

  1. [1]

    Chameleon: Mixed-modal early-fusion foundation models.arXiv preprint arXiv:2405.09818, 2024

    Chameleon Team. Chameleon: Mixed-modal early-fusion foundation models.arXiv preprint arXiv:2405.09818, 2024

    Pith/arXiv arXiv 2024

  2. [2]

    Emerging properties in unified multimodal pretraining.arXiv preprint arXiv:2505.14683, 2025

    Chaorui Deng, Deyao Zhu, Kunchang Li, Chenhui Gou, Feng Li, Zeyu Wang, Shu Zhong, Weihao Yu, Xiaonan Nie, Ziang Song, Guang Shi, and Haoqi Fan. Emerging properties in unified multimodal pretraining.arXiv preprint arXiv:2505.14683, 2025

    Pith/arXiv arXiv 2025

  3. [3]

    Interleaved-modal chain-of-thought.arXiv preprint arXiv:2411.19488, 2024

    Jun Gao, Yongqi Li, Ziqiang Cao, and Wenjie Li. Interleaved-modal chain-of-thought.arXiv preprint arXiv:2411.19488, 2024

    Pith/arXiv arXiv 2024

  4. [4]

    Emu3: Next-token prediction is all you need

    Xinlong Wang, Xiaosong Zhang, Zhengxiong Luo, Quan Sun, Yufeng Cui, Jinsheng Wang, Fan Zhang, Yueze Wang, Zhen Li, Qiying Yu, et al. Emu3: Next-token prediction is all you need. arXiv preprint arXiv:2409.18869, 2024

    Pith/arXiv arXiv 2024

  5. [5]

    Emu3.5: Native multimodal models are world learners.arXiv preprint arXiv:2510.26583, 2025

    Yufeng Cui, Honghao Chen, Haoge Deng, Xu Huang, Xinghang Li, Jirong Liu, Yang Liu, Zhuoyan Luo, Jinsheng Wang, Wenxuan Wang, et al. Emu3.5: Native multimodal models are world learners.arXiv preprint arXiv:2510.26583, 2025

    Pith/arXiv arXiv 2025

  6. [6]

    Visual generation unlocks human-like reasoning through multimodal world models.arXiv preprint arXiv:2601.19834, 2026

    Jialong Wu, Xiaoying Zhang, Hongyi Yuan, Xiangcheng Zhang, Tianhao Huang, Changjing He, Chaoyi Deng, Renrui Zhang, Youbin Wu, and Mingsheng Long. Visual generation unlocks human-like reasoning through multimodal world models.arXiv preprint arXiv:2601.19834, 2026

    arXiv 2026

  7. [7]

    Thinkmorph: Emergent properties in multimodal interleaved chain-of-thought reasoning.arXiv preprint arXiv:2510.27492, 2025

    Jiawei Gu, Yunzhuo Hao, Huichen Will Wang, Linjie Li, Michael Qizhe Shieh, Yejin Choi, Ranjay Krishna, and Yu Cheng. Thinkmorph: Emergent properties in multimodal interleaved chain-of-thought reasoning.arXiv preprint arXiv:2510.27492, 2025

    arXiv 2025

  8. [8]

    Thinking with generated images.arXiv preprint arXiv:2505.22525, 2025

    Ethan Chern, Zhulin Hu, Steffi Chern, Siqi Kou, Jiadi Su, Yan Ma, Zhijie Deng, and Pengfei Liu. Thinking with generated images.arXiv preprint arXiv:2505.22525, 2025

    Pith/arXiv arXiv 2025

  9. [9]

    Deepseek-r1: Incentivizing reasoning capability in LLMs via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in LLMs via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

    Pith/arXiv arXiv 2025

  10. [10]

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

    Pith/arXiv arXiv 2024

  11. [11]

    Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforcement learning.arXiv preprint arXiv:2503.07365, 2025

    Fanqing Meng, Lingxiao Du, Zongkai Liu, Zhixiang Zhou, Quanfeng Lu, Daocheng Fu, Tiancheng Han, Botian Shi, Wenhai Wang, Junjun He, Kaipeng Zhang, Ping Luo, Yu Qiao, Qiaosheng Zhang, and Wenqi Shao. Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforcement learning.arXiv preprint arXiv:2503.07365, 2025

    Pith/arXiv arXiv 2025

  12. [12]

    R1-onevision: Advancing generalized multimodal reasoning through cross-modal formalization.arXiv preprint arXiv:2503.10615, 2025

    Yi Yang, Xiaohui He, Haotian Pan, Xiaojiang Jiang, Yao Deng, Xiaoyang Yang, Hao Lu, Dawei Yin, Fei Rao, Miao Zhu, et al. R1-onevision: Advancing generalized multimodal reasoning through cross-modal formalization.arXiv preprint arXiv:2503.10615, 2025

    Pith/arXiv arXiv 2025

  13. [13]

    COOPER: A unified model for cooperative perception and reasoning in spatial intelligence.arXiv preprint arXiv:2512.04563, 2025

    Zefeng Zhang, Xiangzhao Hao, Hengzhu Tang, Zhenyu Zhang, Jiawei Sheng, Xiaodong Li, Zhenyang Li, Li Gao, Daiting Shi, Dawei Yin, and Tingwen Liu. COOPER: A unified model for cooperative perception and reasoning in spatial intelligence.arXiv preprint arXiv:2512.04563, 2025

    arXiv 2025

  14. [14]

    thinking with images

    Ziwei Zheng, Michael Yang, Jack Hong, Chenxiao Zhao, Guohai Xu, Le Yang, Chao Shen, and Xing Yu. Deepeyes: Incentivizing “thinking with images” via reinforcement learning.arXiv preprint arXiv:2505.14362, 2025

    Pith/arXiv arXiv 2025

  15. [15]

    Training diffusion models with reinforcement learning.arXiv preprint arXiv:2305.13301, 2023

    Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. Training diffusion models with reinforcement learning.arXiv preprint arXiv:2305.13301, 2023

    Pith/arXiv arXiv 2023

  16. [16]

    Dpok: Reinforcement learning for fine-tuning text-to-image diffusion models.Advances in Neural Information Processing Systems, 36:79858–79885, 2023

    Ying Fan, Olivia Watkins, Yuqing Du, Hao Liu, Moonkyung Ryu, Craig Boutilier, Pieter Abbeel, Mohammad Ghavamzadeh, Kangwook Lee, and Kimin Lee. Dpok: Reinforcement learning for fine-tuning text-to-image diffusion models.Advances in Neural Information Processing Systems, 36:79858–79885, 2023. 10

    2023

  17. [17]

    Flow-grpo: Training flow matching models via online RL.arXiv preprint arXiv:2505.05470, 2025

    Jie Liu, Gongye Liu, Jiajun Liang, Yangguang Li, Jiaheng Liu, Xintao Wang, Pengfei Wan, Di Zhang, and Wanli Ouyang. Flow-grpo: Training flow matching models via online RL.arXiv preprint arXiv:2505.05470, 2025

    Pith/arXiv arXiv 2025

  18. [18]

    Diffusionnft: Online diffusion reinforcement with forward process.arXiv preprint arXiv:2509.16117, 2025

    Kaiwen Zheng, Huayu Chen, Haotian Ye, Haoxiang Wang, Qinsheng Zhang, Kai Jiang, Hang Su, Stefano Ermon, Jun Zhu, and Ming-Yu Liu. Diffusionnft: Online diffusion reinforcement with forward process.arXiv preprint arXiv:2509.16117, 2025

    Pith/arXiv arXiv 2025

  19. [19]

    UniGRPO: Unified policy optimization for reasoning-driven visual generation.arXiv preprint arXiv:2603.23500, 2026

    Jie Liu, Zilyu Ye, Linxiao Yuan, Shenhan Zhu, Yu Gao, Jie Wu, Kunchang Li, Xionghui Wang, Xiaonan Nie, Weilin Huang, and Wanli Ouyang. UniGRPO: Unified policy optimization for reasoning-driven visual generation.arXiv preprint arXiv:2603.23500, 2026

    arXiv 2026

  20. [20]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems, volume 35, pages 24824–24837, 2022

    2022

  21. [21]

    Large language models are zero-shot reasoners

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large language models are zero-shot reasoners. InAdvances in Neural Information Processing Systems, volume 35, pages 22199–22213, 2022

    2022

  22. [22]

    Visual sketchpad: Sketching as a visual chain of thought for multimodal language models.Advances in Neural Information Processing Systems, 37:139348–139379, 2024

    Yushi Hu, Weijia Shi, Xingyu Fu, Dan Roth, Mari Ostendorf, Luke Zettlemoyer, Noah A Smith, and Ranjay Krishna. Visual sketchpad: Sketching as a visual chain of thought for multimodal language models.Advances in Neural Information Processing Systems, 37:139348–139379, 2024

    2024

  23. [23]

    Whiteboard-of-thought: Thinking step-by- step across modalities

    Sachit Menon, Richard Zemel, and Carl V ondrick. Whiteboard-of-thought: Thinking step-by- step across modalities. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 20016–20031, 2024

    2024

  24. [24]

    Chain of images for intuitively reasoning.arXiv preprint arXiv:2311.09241, 2023

    Fanxu Meng, Haotong Yang, Yiding Wang, and Muhan Zhang. Chain of images for intuitively reasoning.arXiv preprint arXiv:2311.09241, 2023

    Pith/arXiv arXiv 2023

  25. [25]

    Visual thoughts: A unified perspective of understanding multimodal chain-of-thought

    Zihui Cheng, Qiguang Chen, Xiao Xu, Jiaqi WANG, Weiyun Wang, Hao Fei, Yidong Wang, Alex Jinpeng Wang, Zhi Chen, Wanxiang Che, et al. Visual thoughts: A unified perspective of understanding multimodal chain-of-thought. InAdvances in Neural Information Processing Systems, 2025

    2025

  26. [26]

    Show-o: One single transformer to unify multimodal understanding and generation.arXiv preprint arXiv:2408.12528, 2024

    Jinheng Xie, Weijia Mao, Zechen Bai, David Junhao Zhang, Weihao Wang, Kevin Qinghong Lin, Yuchao Gu, Zhijie Chen, Zhenheng Yang, and Mike Zheng Shou. Show-o: One single transformer to unify multimodal understanding and generation.arXiv preprint arXiv:2408.12528, 2024

    Pith/arXiv arXiv 2024

  27. [27]

    Zebra-cot: A dataset for interleaved vision language reasoning.arXiv preprint arXiv:2507.16746, 2025

    Ang Li, Charles Wang, Deqing Fu, Kaiyu Yue, Zikui Cai, Wang Bill Zhu, Ollie Liu, Peng Guo, Willie Neiswanger, Furong Huang, et al. Zebra-cot: A dataset for interleaved vision language reasoning.arXiv preprint arXiv:2507.16746, 2025

    arXiv 2025

  28. [28]

    Imagine while reasoning in space: Multimodal visualization-of-thought.arXiv preprint arXiv:2501.07542, 2025

    Chengzu Li, Wenshan Wu, Huanyu Zhang, Yan Xia, Shaoguang Mao, Li Dong, Ivan Vuli ´c, and Furu Wei. Imagine while reasoning in space: Multimodal visualization-of-thought.arXiv preprint arXiv:2501.07542, 2025

    Pith/arXiv arXiv 2025

  29. [29]

    Metamorph: Multimodal understanding and generation via instruction tuning

    Shengbang Tong, David Fan, Jiachen Li, Yunyang Xiong, Xinlei Chen, Koustuv Sinha, Michael Rabbat, Yann LeCun, Saining Xie, and Zhuang Liu. Metamorph: Multimodal understanding and generation via instruction tuning. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 17001–17012, 2025

    2025

  30. [30]

    Janus: Decoupling visual encoding for unified multimodal understanding and generation.arXiv preprint arXiv:2410.13848, 2024

    Chengyue Wu, Xiaokang Chen, Zhiyu Wu, Yiyang Ma, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, Chong Ruan, and Ping Luo. Janus: Decoupling visual encoding for unified multimodal understanding and generation.arXiv preprint arXiv:2410.13848, 2024

    Pith/arXiv arXiv 2024

  31. [31]

    Transfusion: Predict the next token and diffuse images with one multi-modal model.arXiv preprint arXiv:2408.11039, 2024

    Chunting Zhou, Lili Yu, Arun Babu, Kushal Tirumala, Michihiro Yasunaga, Leonid Shamis, Jacob Kahn, Xuezhe Ma, Luke Zettlemoyer, and Omer Levy. Transfusion: Predict the next token and diffuse images with one multi-modal model.arXiv preprint arXiv:2408.11039, 2024. 11

    Pith/arXiv arXiv 2024

  32. [32]

    DreamLLM: Synergistic multimodal comprehension and creation

    Runpei Dong, Chunrui Han, Yuang Peng, et al. DreamLLM: Synergistic multimodal comprehension and creation. InInternational Conference on Learning Representations, 2024

    2024

  33. [33]

    Vision-r1: Incentivizing reasoning capability in multimodal large language models.arXiv preprint arXiv:2503.06749, 2025

    Wenxuan Huang, Bohan Jia, Zijie Zhai, Shaosheng Cao, Zheyu Ye, Fei Zhao, Zhe Xu, Xu Tang, Yao Hu, and Shaohui Lin. Vision-r1: Incentivizing reasoning capability in multimodal large language models.arXiv preprint arXiv:2503.06749, 2025

    Pith/arXiv arXiv 2025

  34. [34]

    Beyond the Dirac Delta: Mitigating diversity collapse in reinforcement fine-tuning for versatile image generation.arXiv preprint arXiv:2601.12401, 2026

    Jinmei Liu, Haoru Li, Zhenhong Sun, Chaofeng Chen, Yatao Bian, Bo Wang, Daoyi Dong, Chunlin Chen, and Zhi Wang. Beyond the Dirac Delta: Mitigating diversity collapse in reinforcement fine-tuning for versatile image generation.arXiv preprint arXiv:2601.12401, 2026

    arXiv 2026

  35. [35]

    Dancegrpo: Unleashing grpo on visual generation

    Zeyue Xue, Jie Wu, Yu Gao, Fangyuan Kong, Lingting Zhu, Mengzhao Chen, Zhiheng Liu, Wei Liu, Qiushan Guo, Weilin Huang, et al. Dancegrpo: Unleashing grpo on visual generation. arXiv preprint arXiv:2505.07818, 2025

    Pith/arXiv arXiv 2025

  36. [36]

    X-omni: Reinforcement learning makes discrete autoregressive image generative models great again.arXiv preprint arXiv:2507.22058, 2025

    Zigang Geng, Yibing Wang, Yeyao Ma, Chen Li, Yongming Rao, Shuyang Gu, Zhao Zhong, Qinglin Lu, Han Hu, Xiaosong Zhang, et al. X-omni: Reinforcement learning makes discrete autoregressive image generative models great again.arXiv preprint arXiv:2507.22058, 2025

    Pith/arXiv arXiv 2025

  37. [37]

    Towards unified multimodal interleaved generation via group relative policy optimization

    Ming Nie, Chunwei Wang, Jianhua Han, Hang Xu, and Li Zhang. Towards unified multimodal interleaved generation via group relative policy optimization. InAdvances in Neural Information Processing Systems, 2025

    2025

  38. [38]

    Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    Pith/arXiv arXiv 2022

  39. [39]

    Matterport3d: Learning from rgb-d data in indoor environments.arXiv preprint arXiv:1709.06158, 2017

    Angel Chang, Angela Dai, Thomas Funkhouser, Maciej Halber, Matthias Niessner, Manolis Savva, Shuran Song, Andy Zeng, and Yinda Zhang. Matterport3d: Learning from rgb-d data in indoor environments.arXiv preprint arXiv:1709.06158, 2017

    Pith/arXiv arXiv 2017

  40. [40]

    Realsee3d: A large-scale multi-view rgb-d dataset of indoor scenes (version 1.0), 2025

    Linyuan Li, Yan Wu, Xi Li, Lingli Wang, Tong Rao, Jie Zhou, Cihui Pan, and Xinchen Hui. Realsee3d: A large-scale multi-view rgb-d dataset of indoor scenes (version 1.0), 2025

    2025

  41. [41]

    360+ x: A panoptic multi-modal scene understanding dataset

    Hao Chen, Yuqi Hou, Chenyuan Qu, Irene Testini, Xiaohan Hong, and Jianbo Jiao. 360+ x: A panoptic multi-modal scene understanding dataset. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19373–19382, 2024

    2024

  42. [42]

    Jigsaw-r1: A study of rule-based visual reinforcement learning with jigsaw puzzles.arXiv preprint arXiv:2505.23590, 2025

    Zifu Wang, Junyi Zhu, Bo Tang, Zhiyu Li, Feiyu Xiong, Jiaqian Yu, and Matthew B Blaschko. Jigsaw-r1: A study of rule-based visual reinforcement learning with jigsaw puzzles.arXiv preprint arXiv:2505.23590, 2025

    arXiv 2025

  43. [43]

    Visual cot: Unleashing chain-of-thought reasoning in multi-modal language models.arXiv preprint arXiv:2403.16999, 2, 2024

    Hao Shao, Shengju Qian, Han Xiao, Guanglu Song, Zhuofan Zong, Letian Wang, Yu Liu, and Hongsheng Li. Visual cot: Unleashing chain-of-thought reasoning in multi-modal language models.arXiv preprint arXiv:2403.16999, 2, 2024

    Pith/arXiv arXiv 2024

  44. [44]

    Mmsi-bench: A benchmark for multi-image spatial intelligence.arXiv preprint arXiv:2505.23764, 2025

    Sihan Yang, Runsen Xu, Yiman Xie, Sizhe Yang, Mo Li, Jingli Lin, Chenming Zhu, Xiaochen Chen, Haodong Duan, Xiangyu Yue, et al. Mmsi-bench: A benchmark for multi-image spatial intelligence.arXiv preprint arXiv:2505.23764, 2025

    Pith/arXiv arXiv 2025

  45. [45]

    Sat: Spatial aptitude training for multimodal language models.arXiv preprint arXiv:2412.07755, 3, 2024

    Arijit Ray, Jiafei Duan, Reuben Tan, Dina Bashkirova, Rose Hendrix, Kiana Ehsani, Aniruddha Kembhavi, Bryan A Plummer, Ranjay Krishna, Kuo-Hao Zeng, et al. Sat: Spatial aptitude training for multimodal language models.arXiv preprint arXiv:2412.07755, 3, 2024

    arXiv 2024

  46. [46]

    Eyes wide shut? exploring the visual shortcomings of multimodal llms

    Shengbang Tong, Zhuang Liu, Yuexiang Zhai, Yi Ma, Yann LeCun, and Saining Xie. Eyes wide shut? exploring the visual shortcomings of multimodal llms. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9568–9578, 2024

    2024

  47. [47]

    Cambrian-1: A fully open, vision-centric exploration of multimodal llms.Advances in Neural Information Processing Systems, 37:87310–87356, 2024

    Shengbang Tong, Ellis Brown, Penghao Wu, Sanghyun Woo, Manoj Middepogu, Sai C Akula, Jihan Yang, Shusheng Yang, Adithya Iyer, Xichen Pan, et al. Cambrian-1: A fully open, vision-centric exploration of multimodal llms.Advances in Neural Information Processing Systems, 37:87310–87356, 2024. 12

    2024

  48. [48]

    Blink: Multimodal large language models can see but not perceive

    Xingyu Fu, Yushi Hu, Bangzheng Li, Yu Feng, Haoyu Wang, Xudong Lin, Dan Roth, Noah A Smith, Wei-Chiu Ma, and Ranjay Krishna. Blink: Multimodal large language models can see but not perceive. InEuropean Conference on Computer Vision, pages 148–166. Springer, 2024

    2024

  49. [49]

    V*: Guided visual search as a core mechanism in multimodal llms

    Penghao Wu and Saining Xie. V*: Guided visual search as a core mechanism in multimodal llms. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13084–13094, 2024

    2024

  50. [50]

    Vlmevalkit: An open-source toolkit for evaluating large multi-modality models

    Haodong Duan, Junming Yang, Yuxuan Qiao, Xinyu Fang, Lin Chen, Yuan Liu, Xiaoyi Dong, Yuhang Zang, Pan Zhang, Jiaqi Wang, et al. Vlmevalkit: An open-source toolkit for evaluating large multi-modality models. InProceedings of the 32nd ACM international conference on multimedia, pages 11198–11201, 2024

    2024

  51. [51]

    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

    Pith/arXiv arXiv 2024

  52. [52]

    CLIP-blind pairs

    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, Hang Zhang, Zhibo Yang, Haiyang Xu, and Junyang Lin. Qwen2.5-vl technical report, 2...

    arXiv 2025