Pith.

Computer Vision and Pattern Recognition · arXiv:2611.16517 v1 · submitted 12 Nov 2026 · 178 references

UniSpark: Exploring Hybrid Attention for Multimodal Understanding and Generation

Jiancheng Pan, Yuzheng Liu, Hengyuan Cao, Qiyuan Zhang, Yangyan Li, Biao Gong, Zheng Zhang, Ka Long Cheng

show 2 more authors

Xinyang Wang, Yujun Shen

The paper says

“We present UniSpark-9B, a unified multimodal model that investigates hybrid attention for visual understanding and generation.”

more of the abstract

Built on a Mixture-of-Transformer-Experts backbone with approximately 9B parameters in each branch, UniSpark combines full attention for explicit token interactions with GatedDeltaNet linear attention for compact context propagation. Autoregressive text prediction and visual flow matching jointly support visual question answering, text-to-image generation, instruction-guided editing, and interleaved text-image prediction. In interleaved generation, noisy visual targets are isolated from subsequent predictions, while clean visual representations provide context for later outputs. A blockwise protocol enforces this distinction through attention masks in full-attention layers and recurrent-state retention rules in linear-attention layers. The selected training sources contain 159.7M generation samples, 213.4M understanding samples, and 624.8K interleaved sequences. A four-stage curriculum jointly trains multimodal understanding and visual generation, refining answer accuracy and instruction following while extending image synthesis to 2K. UniSpark scores 2470.3 on MME-S and 0.83 on GenEval. Qualitative examples further illustrate its capabilities in interleaved image generation, video synthesis, and video editing. Together, these results demonstrate that hybrid attention provides an effective backbone for unified multimodal understanding and generation.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

The passes disagreed. Reader: holds up. Skeptic: agrees with the reader. Referee: major revision.

UniSpark is a solid engineering report with a genuine architectural combination, but the central claim about hybrid attention being the effective backbone is correlational, not causal. This line is wrong

3 major objections · 9 minor

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 178 references say.

The paper says, at Abstract, §6.7, b428

The review did not quote this passage.

A classifier read this sentence of the paper as acknowledging the concern (probability 0.90):

“These system-level measurements do not isolate the effect of hybrid attention. 24 (a) Replacing a plate with a terracotta dish Edit instruction: Replace the black plate with a hand-thrown terracotta dish with a slightly uneven rim and a rustic texture. 0.00 s 1.12 s 2.25 s 3.38 s 4.50 s Source Edited (b) Pencil sketch stylization Edit instruction: Make it a pencil sketch style. 0.00 s 1.12 s 2.44 s 3.12 s 4.50 s Source Edited Figure 13:Qualitative examples of video editing.Temporally aligned source and edited frames illustrate object replacement and global stylization.”

Pith says: Objection 1 of 3 · major

The central claim—that hybrid attention provides an effective backbone—is not supported by controlled ablation. The authors acknowledge in §6.7 that 'system-level measurements do not isolate the effect of hybrid attention' (b428). No experiment removes either attention type, varies the layer arrangement, or swaps GatedDeltaNet for a different linear-attention mechanism.

more of objection 1

Confounders such as larger per-branch parameter count (9B vs. 7B of BAGEL), different training data, MoT routing, and the four-stage curriculum can equally explain the observed benchmark scores. Without an ablation study, the evidence for the headline claim is correlational, not causal. This is a load-bearing gap because the title and abstract position hybrid attention as the key innovation.

The paper says, at §3.2, Fig. 5

The review did not quote this passage.

Pith says: Objection 2 of 3 · major

The blockwise public/private state protocol is a clear design contribution, but its benefit over simpler alternatives (e.g., always resetting states without block-level rules, or using only full attention with causal masks) is not experimentally tested. An ablation that compares the full blockwise protocol against a baseline that treats all visual blocks as public (or always private) would isolate the protocol's contribution.

more of objection 2

Without such a test, it is unknown whether the protocol as implemented helps, hurts, or is neutral relative to a simpler approach.

The paper says, at §6.3, Tab. 7

The review did not quote this passage.

Pith says: Objection 3 of 3 · major

On DPG-Bench, UniSpark scores 80.54, substantially below BAGEL (85.07) and LLaDA2.0-Uni (87.76). The paper attributes the overall system's competitiveness to hybrid attention, yet the generation benchmark shows a clear gap. This discrepancy weakens the claim that hybrid attention is generally 'effective' for both understanding and generation. The authors should discuss why hybrid attention would benefit understanding more than generation, or whether the performance gap stems from factors other than attention type (e.g., data mixture, resolution training).

9 minor objections

The paper says, at §3.2, Eq. (2)

Pith says: minor objection 1 of 9

The GatedDeltaNet update equation is presented for a single head. Clarify whether the actual implementation uses multi-head versions and, if so, how heads interact with the shared state.

The paper says, at §4, Table 1 caption

Pith says: minor objection 2 of 9

The caption says 'Source samples before filtering and annotation acceptance'; explain how many samples remain after filtering and how the final 624.8K interleaved sequences relate to the 24.24M source pool.

show 7 more

The paper says, at §4.7, Fig. 9

Pith says: minor objection 3 of 9

The description of online packing mentions a 'soft token target' and 'per-example threshold' but does not specify numerical values for the target or how the threshold is determined. Adding reference values or a sensitivity analysis would improve reproducibility.

The paper says, at §5, Table 3

Pith says: minor objection 4 of 9

The SFT settings are labelled as 'preliminary' but used for the main GenEval and DPG-Bench results. Clarify whether the reported generation scores correspond to the SFT stage with these settings or to a later checkpoint.

The paper says, at §6.1

Pith says: minor objection 5 of 9

The phrase 'UniSpark-9B uses approximately 9B parameters in each of its understanding and generation branches' is repeated nearly verbatim from §3. Consider removing one occurrence for brevity.

The paper says, at §6.2, Table 5

Pith says: minor objection 6 of 9

The MMBench protocol difference is noted but only in the caption. A brief discussion in the text about how the answer extraction difference might affect comparability with Qwen3.5 and SenseNova-U1 would be helpful.

The paper says, at §6.4, Table 8

Pith says: minor objection 7 of 9

The ImgEdit scores are labelled 'reference results' because a different judge was used. Consider including a small-scale inter-judge agreement calibration (e.g., re-run 50 samples with both judges) to bracket the possible bias.

The paper says, at §6.7, Table 9

Pith says: minor objection 8 of 9

The 512×512 latency for UniSpark is 7.27s with 35.19 tokens/s, while BAGEL is 11.76s and 87.09 tokens/s. The tokens/s metric uses different token definitions (BAGEL uses 4096 tokens at 1024^2, UniSpark uses 1024). This makes tokens/s not directly comparable across models; consider reporting only total latency or normalizing by resolution.

The paper says, at General

Pith says: minor objection 9 of 9

The paper contains numerous instances of parenthetical citations without verbs (e.g., 'Kimi Linear adds finer gating through Kimi Delta Attention and combines it with Multi-Head Latent Attention across layers (Kimi Team et al., 2025b).'). Consider integrating the references more naturally.

What would settle it

Train a variant that replaces all GatedDeltaNet layers with full-attention layers (matching total layer count and adjusting dimension to keep parameter count similar). Keep all other training details (model scale, data mixture, MoT routing, curriculum) identical. Evaluate on MME-S, MMBench, GenEval, and DPG-Bench at 512-scale. If scores stay within 2% of the original, hybrid attention is not the primary driver; a significant drop would support its importance.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 13 Nov 2026

  1. Reader holds up confidence high
  2. Skeptic agrees 31 s · 31,291 tokens read, 3,090 written
  3. Editor wrote the verdict above 26 s · 32,159 tokens read, 1,178 written
  4. Referee major revision: 3 major objections, 9 minor 42 s · 31,645 tokens read, 2,060 written

Not run on this paper: the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

UniSpark is a serious engineering effort. The combination of MoT with alternating full-attention and GatedDeltaNet layers is new, and the blockwise protocol for clean vs. noisy visual context is a thoughtful design for interleaved generation. The understanding results are strong—top scores on MME-S, MMBench, MMVP among unified models—and the authors are transparent about their limitations. They list confounders, state that their system-level measurements don't isolate hybrid attention, and note the evaluation protocol differences. That honesty earns credit.

The soft spot is precisely the one the stress-test flags: the paper's headline claim—that hybrid attention provides an effective backbone—is supported only by whole-system benchmark scores. No ablation compares hybrid vs. pure full attention, or varies the layer arrangement. Performance could come from larger per-branch params (9B vs. 7B for BAGEL), richer data, or the MoT routing. The authors themselves acknowledge this in Section 6.7. So the evidence is correlational, not causal. The generation results (DPG-Bench 80.54 vs. BAGEL 85.07) also suggest the hybrid design doesn't help as much on generation quality. The editing evaluation uses a different judge, making comparison with baselines unreliable.

Is that fatal? For a top conference paper, missing the controlled experiment is a load-bearing gap. For a technical report meant to share an architecture and recipe, it's acceptable. The blockwise protocol and hybrid MoT combination are useful for practitioners building unified models. I'd bring this to a reading group focused on multimodal system design, but not for a theory discussion.

Recommendation: This paper deserves a fair peer review if submitted as a conference paper, but with strong expectation that the authors either run an ablation or soften the causal claim in the title and abstract. As a desk editor, I'd send it out because the architecture and honest reporting outweigh the missing control.

The strongest objection

The central claim—that hybrid attention provides an effective backbone for unified multimodal understanding and generation—rests entirely on the full system's competitive benchmark scores. The paper explicitly acknowledges (Section 6.7, b428) that 'system-level measurements do not isolate the effect of hybrid attention.' No ablation removes either attention type or varies the layer arrangement. Consequently, the observed performance could equally be explained by the larger per-branch parameter count (9B vs. 7B for BAGEL), the richer and more curated training data (159.7M generation + 213.4M understanding samples), the MoT expert routing, or the four-stage curriculum. The blockwise protocol and joint objectives are novel but the contribution of hybrid attention per se is untested. Without a controlled experiment that swaps or ablates the GatedDeltaNet layers, the evidence for the headline claim is correlational rather than causal. This is a load-bearing gap because the title and abstract centrally position hybrid attention as the key innovation.

The reader already identified the lack of attention-type ablation as the weakest assumption. I agree that this is unresolved and directly undermines attribution of the central claim. However, the paper presents itself as an exploratory technical report, honestly acknowledges the limitation, and provides a well-documented system with competitive results. The missing ablation is a real weakness but not a fatal flaw given the paper's stated scope; the verdict of ACCEPT with medium correctness risk still fairly represents the evidence. No adjustment is needed.

The referee's report

If the attributed gains are causal, the hybrid attention design (full attention plus GatedDeltaNet with blockwise state rules) offers a practical architecture for unified multimodal models, balancing explicit token interactions with efficient context propagation. The interleaved context-handling protocol and the four-stage training recipe are technically well-specified. The paper's significance is moderate: it scales existing ideas (MoT, GatedDeltaNet) to a large unified model, but the novelty per component is incremental. The main value is as a systems demonstration that competitive results can be achieved with this particular combination of architectural choices and training data.

Pith's reading of the paper (a machine's summary, not the authors')

The paper introduces UniSpark-9B, a single model that handles visual question answering, text-to-image generation, instruction-guided editing, and interleaved text–image prediction. Its central idea is to combine full attention (for explicit token interactions) with GatedDeltaNet linear attention (for efficient context propagation) inside a Mixture-of-Transformer-Experts backbone. The authors design a blockwise information-flow protocol that keeps noisy visual targets private while allowing clean visual representations to condition later outputs. Joint training with autoregressive text prediction and flow matching produces strong benchmark scores and supports high-resolution synthesis up to 2K. The work aims to show that hybrid attention is a practical backbone for unified multimodal models.

What carries the argument. The Hybrid MoT Architecture, which combines token-type expert routing (understanding expert for text and semantic visual features, generation expert for VAE latents) with a layer-wise mixture of full attention (explicit query–key interactions) and GatedDeltaNet linear attention (recurrent matrix state with delta updates). A blockwise information-flow protocol enforces that noisy visual targets (during denoising) remain private and do not update the public recurrent state or attention cache, while clean visual representations (from completed images) do update shared context. Joint training uses autoregressive text cross-entropy and flow-matching on VAE latents.

Core claim. The paper claims that a hybrid attention architecture, alternating full-attention layers and GatedDeltaNet linear-attention layers within an MoT framework, jointly trained on understanding and generation objectives, can achieve state-of-the-art or competitive results across multiple vision-language tasks while also supporting interleaved generation. Specifically, UniSpark-9B reports 2470.3 on MME-S, 86.7 on MMBench, 72.0 on MMVP for understanding, and 0.83 on GenEval for generation, alongside qualitative demonstrations of editing, video generation, and interleaved sequences. The authors assert that the hybrid design provides effective context sharing for both explicit multimodal reasoning and compact recurrent propagation.

The premise it rests on. The reported performance gains are attributed to the hybrid attention design, but the paper acknowledges that system-level measurements (differing model sizes, training data, implementations) do not isolate the effect of hybrid attention from confounding factors.

If the paper is right (Pith's inferences, not the authors')

Other unified multimodal models could adopt similar hybrid attention backbones to balance explicit interactions and efficient long-sequence context.

The blockwise state-retention rule provides a concrete mechanism for reusing generated visual content as context in multi-turn interleaved tasks without exposing intermediate noise.

The four-stage curriculum (foundation, resolution scaling, high-resolution refinement, SFT) offers a recipe for scaling a single model to 2K image generation while preserving understanding performance.

The approach can be extended to video generation and editing by freezing the understanding branch and fine-tuning on video data, as shown in post-SFT experiments.

If the hybrid attention mechanism is the true driver of performance (which the paper acknowledges is not isolated), future work could ablate the contribution of each attention type in controlled settings to guide architectural choices.

The context separation protocol may apply beyond vision-language tasks to any generative system where intermediate noisy states must be shielded from downstream conditioning, such as iterative refinement in audio or time-series generation.

The model's weaker performance on compositional generation (counting, attribute binding) suggests that hybrid attention alone does not resolve these challenges and may require dedicated training strategies or auxiliary losses.

The paper's inference efficiency measurements indicate that the hybrid architecture yields latency benefits at high resolutions (2048×2048) compared to some baselines, which could influence deployment decisions for real-time or resource-constrained applications.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

We present UniSpark-9B, a unified multimodal model that investigates hybrid attention for visual understanding and generation. Built on a Mixture-of-Transformer-Experts backbone with approximately 9B parameters in each branch, UniSpark combines full attention for explicit token interactions with GatedDeltaNet linear attention for compact context propagation. Autoregressive text prediction and visual flow matching jointly support visual question answering, text-to-image generation, instruction-guided editing, and interleaved text-image prediction. In interleaved generation, noisy visual targets are isolated from subsequent predictions, while clean visual representations provide context for later outputs. A blockwise protocol enforces this distinction through attention masks in full-attention layers and recurrent-state retention rules in linear-attention layers. The selected training sources contain 159.7M generation samples, 213.4M understanding samples, and 624.8K interleaved sequences. A four-stage curriculum jointly trains multimodal understanding and visual generation, refining answer accuracy and instruction following while extending image synthesis to 2K. UniSpark scores 2470.3 on MME-S and 0.83 on GenEval. Qualitative examples further illustrate its capabilities in interleaved image generation, video synthesis, and video editing. Together, these results demonstrate that hybrid attention provides an effective backbone for unified multimodal understanding and generation.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-13

The authors' main claim, as the classifier picked it out 1.00

“Together, these results demonstrate that hybrid attention provides an effective backbone for unified multimodal understanding and generation.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.05 · human participants 0.03 · controlled comparison 0.03 · randomized assignment 0.01 · states quantitative result 0.99 · compares with prior results 0.07 · mentions code or data availability 0.03

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
We present UniSpark-9B, a unified multimodal model that investigates hybrid attention for visual understanding and generation. contribution 1.00 0.93 0.07 asserted 0.98
Built on a Mixture-of-Transformer-Experts backbone with approximately 9B parameters in each branch, UniSpark combines full attention for explicit token interactions with GatedDeltaNet linear attention for compact context propagation. method 0.98 0.83 0.04 asserted 1.00
Autoregressive text prediction and visual flow matching jointly support visual question answering, text-to-image generation, instruction-guided editing, and interleaved text-image prediction. method 0.65 0.81 0.05 asserted 1.00
In interleaved generation, noisy visual targets are isolated from subsequent predictions, while clean visual representations provide context for later outputs. method 0.98 0.75 0.03 asserted 1.00
A blockwise protocol enforces this distinction through attention masks in full-attention layers and recurrent-state retention rules in linear-attention layers. method 1.00 0.82 0.03 asserted 1.00
The selected training sources contain 159.7M generation samples, 213.4M understanding samples, and 624.8K interleaved sequences. method 1.00 0.91 0.02 asserted 1.00
A four-stage curriculum jointly trains multimodal understanding and visual generation, refining answer accuracy and instruction following while extending image synthesis to 2K. method 0.97 0.93 0.05 asserted 1.00
UniSpark scores 2470.3 on MME-S and 0.83 on GenEval. result 1.00 0.94 0.06 asserted 1.00
Qualitative examples further illustrate its capabilities in interleaved image generation, video synthesis, and video editing. result 0.93 0.84 0.02 asserted 0.99
Together, these results demonstrate that hybrid attention provides an effective backbone for unified multimodal understanding and generation. interpretation 1.00 0.71 0.94 asserted 1.00

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.90

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 7,570 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (13)
Figure 1 of the paper
Figure 1. Multimodal understanding, generation, and editing performance. 1 arXiv:2611.16517v1 [cs.CV] 12 Nov 2026
Figure 2 of the paper
Figure 2. Text-to-image generation. 2
Figure 3 of the paper
Figure 3. Image editing and interleaved generation. Editing pairs: source (left), result (right). 3
Figure 4 of the paper
Figure 4. Hybrid MoT Architecture of UniSpark. Token-type routing selects understanding or generation parameters, while full-attention and recurrent linear-attention layers exchange shared multimodal context for text prediction, image generation, editing, and interleaved outputs. where i indexes tokens, αi controls state decay, and βi controls the delta update. Understanding and generation tokens use their respective parameters to compute these quantities but share the recurrent state and short convolution history along the permitted multimodal history. Switching experts therefore does not…
Figure 5 of the paper
Figure 5. Information flow under hybrid attention. Visibility masks and recurrent-state rules preserve the same public/private boundary, allowing clean context reuse without propagating noisy targets. 4 Data Preparation UniSpark combines data for multimodal understanding, image generation, image editing, text-only learning, and interleaved prediction during its main training phase. We organize these sources by their supervision structure, while adjusting resolution and sampling separately across training stages. The curriculum com￾prises Joint Multimodal Foundation, Progressive Resolution…
Figure 6 of the paper
Figure 6. Training-data composition and task examples. Source counts and examples summarize the task coverage of joint training, from multimodal understanding to image generation, editing, and interleaved prediction. Data scale and coverage. The selected data sources for main training contain 159.66M generation samples and 213.39M understanding samples. The latter comprise 99.98M captioning samples, 97.90M visual knowl￾edge QA samples, 9.99M OCR samples, 4.51M grounding samples, and 1.00M GUI samples. These sources also include 97.90M text-only samples, 0.50M editing examples and 624.8K…
Figure 7 of the paper
Figure 7. Video-data composition and examples. Generation sources (a) and aligned editing pairs (b) provide 24.24M and 3.29M samples, respectively, for post-SFT video adaptation. Counts refer to indexed samples; byte-identical Ditto shards are counted once. VCG accounts for 89.8% of the video generation samples, complemented by Aishi and other sources. The video editing data cover global edits (2.07M), local edits (557.4K), and simulation-to-real (Sim2Real) trans￾formations (669.5K), with temporally aligned source–target clips. These data support exploratory post-SFT fine-tuning; their…
Figure 8 of the paper
Figure 8. Video-based interleaved data construction. Event-selected video frames are paired with scene descriptions and transition instructions to form grounded text–image training sequences. Quality screening and annotation validation. The planned screening uses resolution, duration, clarity, exposure, aesthetics, motion, shot boundaries, and overlays to identify unsuitable clips and split clips longer than 30 seconds. Before use, annotations must be checked for correct format and timing: valid role alternation, one target image per assistant turn, valid media references, strictly increasing…
Figure 9 of the paper
Figure 9. Multimodal sequence packing. Online packing combines task-specific samples within a token budget while preserving sample isolation and the public/private boundaries required by hybrid attention. Packing concatenates multiple examples while retaining their individual sequence boundaries and within￾example block structure. Text labels and visual targets are gathered through separate loss indices, so conditioning content can share the input sequence without receiving a prediction loss. For interleaved samples, the complete sequence of transitions and images remains within the same…
Figure 10 of the paper
Figure 10. Data sampling across training stages. Representative task weights are shown for three pre￾SFT stages; asterisks mark mandatory groups. These weights do not represent observed sample or token proportions. Stage Name Generation scale Primary objective I Joint Multimodal Foundation 256-scale Establish joint text, visual understanding, and latent￾generation modeling. II Progressive Resolution Scaling 512-scale Expand spatial detail while continuing multimodal supervision. III High-Resolution Refinement 1024-scale to 2K Extend visual generation to longer target sequences and finer…
Figure 11 of the paper
Figure 11. Text-only generation (top) and first-frame-conditioned editing (bottom). Generation predicts F1–F4 jointly, whereas editing conditions on F1 to predict F2–F4, illustrating visual continuity under two conditioning settings. multimodal prefill, and text decoding. Decoding throughput is 127/(t128 − t1), where ti is a CUDA event recorded after the output projection for prediction i. First-logit latency measures request start to t1, before token selection; it is not user-visible time to first token. Tab. 10 places UniSpark between BAGEL and SenseNova-U1 in decoding throughput and total…
Figure 12 of the paper
Figure 12. Qualitative examples of video generation. Chronological frames illustrate text-conditioned scene evolution, including camera movement, waves, and subtle human motion. The two tables distinguish generation speed from autoregressive decoding speed: UniSpark reduces gen￾eration latency relative to BAGEL at the shared resolutions, whereas BAGEL remains faster and uses less allocated memory for understanding. Matched workloads control sequence lengths, but model sizes, native tokenization, vision encoders, and implementations still differ. These system-level measurements do not isolate…
Figure 13 of the paper
Figure 13. Qualitative examples of video editing. Temporally aligned source and edited frames illustrate object replacement and global stylization. Resolution Model # LLM Params Image tokens/s↑ Median latency (s)↓ P95 latency (s)↓ Peak allocated memory (GB)↓ BAGEL (2025) 7B MoT 87.09 11.76 11.77 30.47 512 × 512 SenseNova-U1 (2026) 8B MoT 39.80 6.43 6.52 35.49 UniSpark 9B MoT 35.19 7.27 7.36 37.39 BAGEL (2025) 7B MoT 89.44 45.80 45.81 33.29 1024 × 1024 SenseNova-U1 (2026) 8B MoT 73.90 13.86 13.91 35.84 UniSpark 9B MoT 72.33 14.16 14.21 42.47 BAGEL (2025) 7B MoT – – – – 2048 × 2048 SenseNova-U1…

Figures are the authors', reproduced from the paper.

References (178)
  1. [1] Qingyan Bai, Qiuyu Wang, Hao Ouyang, Yue Yu, Hanlin Wang, Wen Wang, Ka Leong Cheng, Shuailei Ma, Yanhong Zeng, Zichen Liu, Yinghao Xu, Yujun Shen, and Qifeng Chen. Scaling instruction-based video editing with a high-quality synthetic dataset. In Proceedings of the IEEE/CVF Con... on Pith
  2. [2] Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, et al. Qwen2.5-VL technical report. arXiv preprint arXiv:2502.13923, 2025. doi:10.48550/arXiv.2502.13923. URL https://arxiv.org/abs/2502.13923 on PithPith's reviewarXivdoi:10.48550/arxiv.2502.13923
  3. [3] Black Forest Labs . FLUX.1 [dev] Model Card . https://huggingface.co/black-forest-labs/FLUX.1-dev, 2024 on Pith
  4. [4] Black Forest Labs , Stephen Batifol, Andreas Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dockhorn, Jack English, Zion English, Patrick Esser, et al. FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space . arXiv preprint arX... on PithPith's reviewarXiv
  5. [5] Chameleon Team . Chameleon: Mixed-Modal Early-Fusion Foundation Models . arXiv preprint arXiv:2405.09818, 2024. URL https://arxiv.org/abs/2405.09818 on PithPith's reviewarXiv
  6. [6] Junsong Chen, Jincheng Yu, Yitong Li, Shuchen Xue, Haozhe Liu, Jingyu Xin, Yuyang Zhao, Tian Ye, Zhangjie Wu, Zian Wang, Daquan Zhou, Ping Luo, Song Han, and Enze Xie. SANA-Video 2.0 : Hybrid linear attention with attention residuals for efficient video generation. arXiv prepr... on PithPith's reviewarXivdoi:10.48550/arxiv.2607.21553
  7. [8] Xiaokang Chen, Zhiyu Wu, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, and Chong Ruan. Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling . arXiv preprint arXiv:2501.17811, 2025. URL https://arxiv.org/abs/2501.17811 on PithPith's reviewarXiv
  8. [9] 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. UR... on PithPith's reviewarXiv
  9. [11] Haiwen Diao, Penghao Wu, Hanming Deng, Jiahao Wang, Shihao Bai, Silei Wu, Weichen Fan, Wenjie Ye, Wenwen Tong, Xiangyu Fan, Yan Li, Yubo Wang, Zhijie Cao, Zhiqian Lin, Zhitao Yang, Zhongang Cai, Yuwei Niu, Yue Zhu, Bo Liu, Chengguang Lv, Haojia Yu, Haozhe Xie, Hongli Wang, Jia... on PithPith's reviewarXivdoi:10.48550/arxiv.2605.12500
  10. [13] Yu Gao, Lixue Gong, Qiushan Guo, Xiaoxia Hou, Zhichao Lai, Fanshi Li, Liang Li, Xiaochen Lian, Chao Liao, Liyang Liu, et al. Seedream 3.0 Technical Report . arXiv preprint arXiv:2504.11346, 2025. URL https://arxiv.org/abs/2504.11346 on PithPith's reviewarXiv
  11. [14] Yuying Ge, Sijie Zhao, Jinguo Zhu, Yixiao Ge, Kun Yi, Lin Song, Chen Li, Xiaohan Ding, and Ying Shan. SEED-X: Multimodal Models with Unified Multi-granularity Comprehension and Generation . arXiv preprint arXiv:2404.14396, 2024. URL https://arxiv.org/abs/2404.14396 on PithPith's reviewarXiv
  12. [15] Dhruba Ghosh, Hanna Hajishirzi, and Ludwig Schmidt. GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignment . arXiv preprint arXiv:2310.11513, 2023. URL https://arxiv.org/abs/2310.11513 on PithPith's reviewarXiv
  13. [16] Xiaoyang Han, Jianhua Li, Kewang Deng, Zukai Chen, Xuanke Shi, Sihan Wang, Boxuan Li, Linyan Wang, Siyi Xie, Xin You, Jinsheng Quan, Zhongang Cai, Haiwen Diao, Ziwei Liu, Lei Yang, Dahua Lin, and Quan Wang. Vision as Unified Multimodal Generation . arXiv preprint arXiv:2607.06... on PithPith's reviewarXivdoi:10.48550/arxiv.2607.06560
  14. [17] Ali Hatamizadeh, Yejin Choi, and Jan Kautz. Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention . arXiv preprint arXiv:2605.22791, 2026. doi:10.48550/arXiv.2605.22791. URL https://arxiv.org/abs/2605.22791 on PithPith's reviewarXivdoi:10.48550/arxiv.2605.22791
  15. [18] Xiwei Hu, Rui Wang, Yixiao Fang, Bin Fu, Pei Cheng, and Gang Yu. ELLA: Equip Diffusion Models with LLM for Enhanced Semantic Alignment . arXiv preprint arXiv:2403.05135, 2024. URL https://arxiv.org/abs/2403.05135 on PithPith's reviewarXiv
  16. [19] Inclusion AI , Tiwei Bie, Haoxing Chen, Tieyuan Chen, Zhenglin Cheng, Long Cui, Kai Gan, Zhicheng Huang, Zhenzhong Lan, Haoquan Li, et al. LLaDA2.0-Uni: Unifying Multimodal Understanding and Generation with Diffusion Large Language Model . arXiv preprint arXiv:2604.20796, 2026... on PithPith's reviewarXiv
  17. [20] Angelos Katharopoulos, Apoorv Vyas, Nikolaos Pappas, and Fran c ois Fleuret. Transformers are RNN s: Fast autoregressive transformers with linear attention. In Proceedings of the 37th International Conference on Machine Learning, volume 119, pp.\ 5156--5165, 2020. URL https://... on Pith
  18. [21] Kimi Team , Angang Du, Bohong Yin, Bowei Xing, Bowen Qu, Bowen Wang, Cheng Chen, Chenlin Zhang, Chenzhuang Du, Chu Wei, et al. Kimi-VL Technical Report . arXiv preprint arXiv:2504.07491, 2025 a . URL https://arxiv.org/abs/2504.07491 on PithPith's reviewarXiv
  19. [23] Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, et al. LLaVA-OneVision: Easy Visual Task Transfer . arXiv preprint arXiv:2408.03326, 2024. URL https://arxiv.org/abs/2408.03326 on PithPith's reviewarXiv
  20. [24] Ruibang Li, Guan Luo, Yiwei Zhang, Jin Gao, Bing Li, and Weiming Hu. SoLA-Vision : Fine-grained layer-wise linear softmax hybrid attention. arXiv preprint arXiv:2601.11164, 2026. doi:10.48550/arXiv.2601.11164. URL https://arxiv.org/abs/2601.11164 on Pithdoi:10.48550/arxiv.2601.11164
show 80 more
  1. [26] Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling. In International Conference on Learning Representations, 2023. URL https://arxiv.org/abs/2210.02747 on PithPith's reviewarXiv
  2. [27] Shiyu Liu, Yucheng Han, Peng Xing, Fukun Yin, Rui Wang, Wei Cheng, Jiaqi Liao, Yingming Wang, Honghao Fu, Chunrui Han, et al. Step1X-Edit: A Practical Framework for General Image Editing . arXiv preprint arXiv:2504.17761, 2025. URL https://arxiv.org/abs/2504.17761 on PithPith's reviewarXiv
  3. [28] Yuan Liu, Haodong Duan, Yuanhan Zhang, Bo Li, Songyang Zhang, Wangbo Zhao, Yike Yuan, Jiaqi Wang, Conghui He, Ziwei Liu, et al. MMBench: Is Your Multi-Modal Model an All-around Player? In Proceedings of the European Conference on Computer Vision (ECCV), 2024. URL https://arxiv... on PithPith's reviewarXiv
  4. [29] Zhiheng Liu, Weiming Ren, Xiaoke Huang, Shoufa Chen, Tianhong Li, Mengzhao Chen, Yatai Ji, Sen He, Jonas Schult, Belinda Zeng, Tao Xiang, Wenhu Chen, Ping Luo, Luke Zettlemoyer, and Yuren Cong. Tuna-2: Pixel Embeddings Beat Vision Encoders for Multimodal Understanding and Gene... on PithPith's reviewarXivdoi:10.48550/arxiv.2604.24763
  5. [30] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In International Conference on Learning Representations, 2019. URL https://arxiv.org/abs/1711.05101 on PithPith's reviewarXiv
  6. [31] Pan Lu, Hritik Bansal, Tony Xia, Jiacheng Liu, Chunyuan Li, Hannaneh Hajishirzi, Hao Cheng, Kai-Wei Chang, Michel Galley, and Jianfeng Gao. MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts . arXiv preprint arXiv:2310.02255, 2023. URL https:/... on PithPith's reviewarXiv
  7. [32] Meituan LongCat Team , Hanghang Ma, Haoxian Tan, Jiale Huang, Junqiang Wu, Jun-Yan He, Lishuai Gao, Songlin Xiao, Xiaoming Wei, Xiaoqi Ma, et al. LongCat-Image Technical Report . arXiv preprint arXiv:2512.07584, 2025. URL https://arxiv.org/abs/2512.07584 on PithPith's reviewarXiv
  8. [33] Boyuan Meng, Peihua Bao, Hong Liu, Xiaowei Zhu, Chao Wang, Gen Li, and Zhenxuan Pan. HARTS: Efficient Agentic Reinforcement Learning for Hybrid-Attention Models over Arbitrary Rollout Trees . arXiv preprint arXiv:2608.28158, 2026. doi:10.48550/arXiv.2608.28158. URL https://arx... on PithPith's reviewarXivdoi:10.48550/arxiv.2608.28158
  9. [34] William Merrill, Yanhong Li, Tyler Romero, Anej Svete, Caia Costello, Pradeep Dasigi, Dirk Groeneveld, David Heineman, Bailey Kuehl, Nathan Lambert, Chuan Li, Kyle Lo, Saumya Malik, DJ Matusz, Benjamin Minixhofer, Jacob Morrison, Luca Soldaini, Finbarr Timbers, Pete Walsh, Noa... on PithPith's reviewarXivdoi:10.48550/arxiv.2604.03444
  10. [35] OpenGVLab . InternVL2 Model Card . https://huggingface.co/OpenGVLab/InternVL2-8B, 2024 on Pith
  11. [36] Xichen Pan, Satya Narayan Shukla, Aashu Singh, Zhuokai Zhao, Shlok Kumar Mishra, Jialiang Wang, Zhiyang Xu, Jiuhai Chen, Kunpeng Li, Felix Juefei-Xu, et al. Transfer between Modalities with MetaQueries . arXiv preprint arXiv:2504.06256, 2025. URL https://arxiv.org/abs/2504.06256 on PithPith's reviewarXiv
  12. [37] Wujian Peng, Lingchen Meng, Yuxuan Cai, Xianwei Zhuang, Yuhuan Yang, Rongyao Fang, Chenfei Wu, Junyang Lin, Zuxuan Wu, and Shuai Bai. Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer is Key to Unification . arXiv preprint arXiv:2606.18249, 2026. ... on PithPith's reviewarXivdoi:10.48550/arxiv.2606.18249
  13. [38] Liao Qu, Huichao Zhang, Yiheng Liu, Xu Wang, Yi Jiang, Yiming Gao, Hu Ye, Daniel K. Du, Zehuan Yuan, and Xinglong Wu. TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation . arXiv preprint arXiv:2412.03069, 2024. URL https://arxiv.org/abs/2412.03069 on PithPith's reviewarXiv
  14. [39] Qwen Team . Qwen-Image-Edit Model Card . https://huggingface.co/Qwen/Qwen-Image-Edit, 2025 on Pith
  15. [40] Qwen Team . Qwen3.5-9B model card. Hugging Face, 2026. URL https://huggingface.co/Qwen/Qwen3.5-9B. Accessed September 6, 2026 on Pith
  16. [41] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj \"o rn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022. URL https://arxiv.org/abs/2112.10752 on PithPith's reviewarXiv
  17. [42] Weijia Shi, Xiaochuang Han, Chunting Zhou, Weixin Liang, Xi Victoria Lin, Luke Zettlemoyer, and Lili Yu. LMFusion: Adapting Pretrained Language Models for Multimodal Generation . arXiv preprint arXiv:2412.15188, 2024. URL https://arxiv.org/abs/2412.15188 on PithPith's reviewarXiv
  18. [43] Zhentao Tan, Wei Chen, Jingyi Shen, Yao Liu, Xu Shen, Yue Wu, and Jieping Ye. HydraHead: From Head-Level Functional Heterogeneity to Specialized Attention Hybridization . arXiv preprint arXiv:2606.20097, 2026. doi:10.48550/arXiv.2606.20097. URL https://arxiv.org/abs/2606.20097 on PithPith's reviewarXivdoi:10.48550/arxiv.2606.20097
  19. [44] Tencent Hunyuan Foundation Model Team . HunyuanImage 3.0 Technical Report . arXiv preprint arXiv:2509.23951, 2025. URL https://arxiv.org/abs/2509.23951 on PithPith's reviewarXiv
  20. [45] Changyao Tian, Danni Yang, Guanzhou Chen, Erfei Cui, Zhaokai Wang, Yuchen Duan, Penghao Yin, Sitao Chen, Ganlin Yang, Mingxin Liu, et al. InternVL-U: Democratizing Unified Multimodal Models for Understanding, Reasoning, Generation and Editing . arXiv preprint arXiv:2603.09877,... on PitharXiv
  21. [46] Shengbang Tong, David Fan, Jiachen Zhu, Yunyang Xiong, Xinlei Chen, Koustuv Sinha, Michael Rabbat, Yann LeCun, Saining Xie, and Zhuang Liu. MetaMorph: Multimodal Understanding and Generation via Instruction Tuning . arXiv preprint arXiv:2412.14164, 2024 a . URL https://arxiv.o... on PithPith's reviewarXiv
  22. [48] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, 2017. URL https://arxiv.org/abs/1706.03762 on PithPith's reviewarXiv
  23. [49] Wan Team . Wan2.2: Open and Advanced Large-Scale Video Generative Models . https://github.com/Wan-Video/Wan2.2, 2025 on Pith
  24. [50] Chunwei Wang, Guansong Lu, Junwei Yang, Runhui Huang, Jianhua Han, Lu Hou, Wei Zhang, and Hang Xu. ILLUME: Illuminating Your LLMs to See, Draw, and Self-Enhance . arXiv preprint arXiv:2412.06673, 2024 a . URL https://arxiv.org/abs/2412.06673 on PithPith's reviewarXiv
  25. [51] Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, et al. Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution . arXiv preprint arXiv:2409.12191, 2024 b . URL https://arxiv.org/... on PithPith's reviewarXiv
  26. [52] 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 c . URL https://arxiv.org/abs/2409.18869 on PithPith's reviewarXiv
  27. [53] Ronald J. Williams and David Zipser. A learning algorithm for continually running fully recurrent neural networks. Neural Computation, 1 0 (2): 0 270--280, 1989. doi:10.1162/neco.1989.1.2.270 on Pithdoi:10.1162/neco.1989.1.2.270
  28. [54] Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng-ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, et al. Qwen-Image Technical Report . arXiv preprint arXiv:2508.02324, 2025 a . URL https://arxiv.org/abs/2508.02324 on PithPith's reviewarXiv
  29. [55] Chengyue Wu, Xiaokang Chen, Zhiyu Wu, Yiyang Ma, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, Chong Ruan, et al. Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation . arXiv preprint arXiv:2410.13848, 2024 a . URL https://arxiv.o... on PithPith's reviewarXiv
  30. [56] Chenyuan Wu, Pengfei Zheng, Ruiran Yan, Shitao Xiao, Xin Luo, Yueze Wang, Wanli Li, Xiyan Jiang, Yexin Liu, Junjie Zhou, et al. OmniGen2: Towards Instruction-Aligned Multimodal Generation . arXiv preprint arXiv:2506.18871, 2025 b . URL https://arxiv.org/abs/2506.18871 on PithPith's reviewarXiv
  31. [57] Penghao Wu, Haiwen Diao, Weichen Fan, Lewei Lu, Dahua Lin, and Ziwei Liu. Uncovering Understanding-Generation Synergy in Native Unified Multimodal Models: From Representation, Task to System . arXiv preprint arXiv:2609.01607, 2026. doi:10.48550/arXiv.2609.01607. URL https://ar... on PithPith's reviewarXivdoi:10.48550/arxiv.2609.01607
  32. [58] Yecheng Wu, Zhuoyang Zhang, Junyu Chen, Haotian Tang, Dacheng Li, Yunhao Fang, Ligeng Zhu, Enze Xie, Hongxu Yin, Li Yi, et al. VILA-U: a Unified Foundation Model Integrating Visual Understanding and Generation . arXiv preprint arXiv:2409.04429, 2024 b . URL https://arxiv.org/a... on PithPith's reviewarXiv
  33. [59] Zhiyu Wu, Xiaokang Chen, Zizheng Pan, Xingchao Liu, Wen Liu, Damai Dai, Huazuo Gao, Yiyang Ma, Chengyue Wu, Bingxuan Wang, et al. DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding . arXiv preprint arXiv:2412.10302, 2024 c . URL https... on PithPith's reviewarXiv
  34. [60] 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 a . doi... on PithPith's reviewarXivdoi:10.48550/arxiv.2408.12528
  35. [61] Jinheng Xie, Zhenheng Yang, and Mike Zheng Shou. Show-o2 : Improved native unified multimodal models. arXiv preprint arXiv:2506.15564, 2025. doi:10.48550/arXiv.2506.15564. URL https://arxiv.org/abs/2506.15564 on PithPith's reviewarXivdoi:10.48550/arxiv.2506.15564
  36. [62] Rongchang Xie, Chen Du, Ping Song, and Chang Liu. MUSE-VL: Modeling Unified VLM through Semantic Discrete Encoding . arXiv preprint arXiv:2411.17762, 2024 b . URL https://arxiv.org/abs/2411.17762 on PithPith's reviewarXiv
  37. [63] Yi Xin, Qi Qin, Siqi Luo, Kaiwen Zhu, Juncheng Yan, Yan Tai, Jiayi Lei, Yuewen Cao, Keqi Wang, Yibin Wang, et al. Lumina-DiMOO: An Omni Diffusion Large Language Model for Multi-Modal Generation and Understanding . arXiv preprint arXiv:2510.06308, 2025 a . URL https://arxiv.org... on PitharXiv
  38. [64] Yi Xin, Juncheng Yan, Qi Qin, Zhen Li, Dongyang Liu, Shicheng Li, Victor Shea-Jay Huang, Yupeng Zhou, Renrui Zhang, Le Zhuo, et al. Lumina-mGPT 2.0: Stand-Alone AutoRegressive Image Modeling . arXiv preprint arXiv:2507.17801, 2025 b . URL https://arxiv.org/abs/2507.17801 on PithPith's reviewarXiv
  39. [65] Le Xue, Manli Shu, Anas Awadalla, Jun Wang, An Yan, Senthil Purushwalkam, Honglu Zhou, Viraj Prabhu, Yutong Dai, Michael S Ryoo, et al. xGen-MM (BLIP-3): A Family of Open Large Multimodal Models . arXiv preprint arXiv:2408.08872, 2024. URL https://arxiv.org/abs/2408.08872 on PitharXiv
  40. [66] Ceyuan Yang, Zhijie Lin, Yang Zhao, Fei Xiao, Hao He, Qi Zhao, Chaorui Deng, Kunchang Li, Zihan Ding, Yuwei Guo, Fuyun Wang, Fangqi Zhu, Xiaonan Nie, Shenhan Zhu, Shanchuan Lin, Hongsheng Li, Weilin Huang, Guang Shi, and Haoqi Fan. Context Unrolling in Omni Models . arXiv prep... on PithPith's reviewarXivdoi:10.48550/arxiv.2604.21921
  41. [67] Ling Yang, Ye Tian, Bowen Li, Xinchen Zhang, Ke Shen, Yunhai Tong, and Mengdi Wang. MMaDA: Multimodal Large Diffusion Language Models . arXiv preprint arXiv:2505.15809, 2025 a . URL https://arxiv.org/abs/2505.15809 on PithPith's reviewarXiv
  42. [70] Zebin You, Xiaolu Zhang, Jun Zhou, Chongxuan Li, and Ji-Rong Wen. LLaDA-o: An Effective and Length-Adaptive Omni Diffusion Model . arXiv preprint arXiv:2603.01068, 2026. URL https://arxiv.org/abs/2603.01068 on PitharXiv
  43. [71] Weihao Yu, Zhengyuan Yang, Linjie Li, Jianfeng Wang, Kevin Lin, Zicheng Liu, Xinchao Wang, and Lijuan Wang. MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilities . In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedin... on Pith
  44. [72] Xiang Yue, Yuansheng Ni, Kai Zhang, Tianyu Zheng, et al. MMMU: A Massive Multi-Discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI . In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. URL https://openaccess... on Pith
  45. [73] Z-Image Team , Huanqia Cai, Sihan Cao, Ruoyi Du, Peng Gao, Steven Hoi, Zhaohui Hou, Shijie Huang, Dengyang Jiang, Xin Jin, Liangchen Li, et al. Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer . arXiv preprint arXiv:2511.22699, 2... on PithPith's reviewarXiv
  46. [74] Huichao Zhang, Liao Qu, Yiheng Liu, Hang Chen, Yangyang Song, Yongsheng Dong, Shikun Sun, Xian Li, Xu Wang, Yi Jiang, et al. NextFlow: Unified Sequential Modeling Activates Multimodal Understanding and Generation . arXiv preprint arXiv:2601.02204, 2026. URL https://arxiv.org/a... on PitharXiv
  47. [75] Yanli Zhao, Andrew Gu, Rohan Varma, Liang Luo, Chien-Chin Huang, Min Xu, Less Wright, Hamid Shojanazeri, Myle Ott, Sam Shleifer, et al. PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel . arXiv preprint arXiv:2304.11277, 2023. URL https://arxiv.org/abs/2304.11277 on PithPith's reviewarXiv
  48. [76] 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. doi:10.4855... on PithPith's reviewarXivdoi:10.48550/arxiv.2408.11039
  49. [77] Yubo Zhu, Zhehan Kan, Jingyi Yang, Miaolin Chen, Jinbo Xing, Kai Zhu, Zijian Wang, Sheng Zhong, and Wei Tong. When Does Visual Generation Help Visual Understanding in Unified Multimodal Models? arXiv preprint arXiv:2608.22174, 2026. doi:10.48550/arXiv.2608.22174. URL https://a... on PithPith's reviewarXivdoi:10.48550/arxiv.2608.22174
  50. [78] Shuai Bai and Yuxuan Cai and Ruizhe Chen and Keqin Chen and Xionghui Chen and Zesen Cheng and Lianghao Deng and Wei Ding and Chang Gao and Chunjiang Ge and others , title =. 2025 , journal = on Pith
  51. [79] 2024 , howpublished = on Pith
  52. [80] 2026 , howpublished = on Pith
  53. [81] 2025 , howpublished = on Pith
  54. [82] Qi Cai and Jingwen Chen and Yang Chen and Yehao Li and Fuchen Long and Yingwei Pan and Zhaofan Qiu and Yiheng Zhang and Fengbin Gao and Peihan Xu and others , title =. 2025 , journal = on Pith
  55. [83] Shuo Cao and Nan Ma and Jiayang Li and Xiaohui Li and Lihao Shao and Kaiwen Zhu and Yu Zhou and Yuandong Pu and Jiarui Wu and Jiaquan Wang and Bo Qu and Wenhai Wang and Yu Qiao and Dajuin Yao and Yihao Liu , title =. 2025 , journal = on Pith
  56. [84] Guoxuan Chen and Chufeng Xiao and Haoran Yang and Siyue Xie and Binxiao Huang and Ming Zhang and Cheuk Him Chau and Xinyu Fu and Yingzhao Lian and others , title =. 2026 , journal = on Pith
  57. [85] Jiuhai Chen and Zhiyang Xu and Xichen Pan and Yushi Hu and Can Qin and Tom Goldstein and Lifu Huang and Tianyi Zhou and Saining Xie and Silvio Savarese and others , title =. 2025 , journal = on Pith
  58. [86] Lin Chen and Jinsong Li and Xiaoyi Dong and Pan Zhang and Yuhang Zang and Zehui Chen and Haodong Duan and Jiaqi Wang and Yu Qiao and Dahua Lin and others , title =. 2024 , journal = on Pith
  59. [87] Xiaokang Chen and Zhiyu Wu and Xingchao Liu and Zizheng Pan and Wen Liu and Zhenda Xie and Xingkai Yu and Chong Ruan , title =. 2025 , journal = on Pith
  60. [88] Xianfu Cheng and Wei Zhang and Shiwei Zhang and Jian Yang and Xiangyuan Guan and Xianjie Wu and Xiang Li and Ge Zhang and Jiaheng Liu and Yuying Mai and others , title =. 2025 , booktitle = on Pith
  61. [89] Zhenglin Cheng and Peng Sun and Jianguo Li and Tao Lin , title =. 2026 , booktitle = on Pith
  62. [90] Chaorui Deng and Deyao Zhu and Kunchang Li and Chenhui Gou and Feng Li and Zeyu Wang and Shu Zhong and Weihao Yu and Xiaonan Nie and Ziang Song and others , title =. 2025 , journal =. doi:10.48550/arXiv.2505.14683 , url = on PithPith's reviewarXivdoi:10.48550/arxiv.2505.14683
  63. [91] Songlin Yang and Jan Kautz and Ali Hatamizadeh , title =. International Conference on Learning Representations , year =. doi:10.48550/arXiv.2412.06464 , url = on PithPith's reviewarXivdoi:10.48550/arxiv.2412.06464
  64. [92] Patrick Esser and Sumith Kulal and Andreas Blattmann and Rahim Entezari and Jonas M. Scaling rectified flow transformers for high-resolution image synthesis , year =. Proceedings of the 41st International Conference on Machine Learning , url = on Pith
  65. [93] Yu Gao and Lixue Gong and Qiushan Guo and Xiaoxia Hou and Zhichao Lai and Fanshi Li and Liang Li and Xiaochen Lian and Chao Liao and Liyang Liu and others , title =. 2025 , journal = on Pith
  66. [94] Zigang Geng and Yibing Wang and Yeyao Ma and Chen Li and Yongming Rao and Shuyang Gu and Zhao Zhong and Qinglin Lu and Han Hu and Xiaosong Zhang and others , title =. 2025 , journal = on Pith
  67. [95] Dhruba Ghosh and Hanna Hajishirzi and Ludwig Schmidt , title =. 2023 , journal = on Pith
  68. [96] Tianrui Guan and Fuxiao Liu and Xiyang Wu and Ruiqi Xian and Zongxia Li and Xiaoyu Liu and Xijun Wang and Lichang Chen and Furong Huang and Yaser Yacoob and others , title =. 2024 , booktitle = on Pith
  69. [97] Junlin Han and Shengbang Tong and David Fan and Minghao Chen and Philip Torr and Filippos Kokkinos and Mike Lewis , title =. 2026 , journal = on Pith
  70. [98] Jack Hessel and Ari Holtzman and Maxwell Forbes and Le Bras, Ronan and Yejin Choi , title =. 2021 , booktitle = on Pith
  71. [99] Xiwei Hu and Rui Wang and Yixiao Fang and Bin Fu and Pei Cheng and Gang Yu , title =. 2024 , journal = on Pith
  72. [100] Pavel Izmailov and Dmitrii Podoprikhin and Timur Garipov and Dmitry Vetrov and Andrew Gordon Wilson , title =. 2018 , booktitle = on Pith
  73. [101] Aniruddha Kembhavi and Mike Salvato and Eric Kolve and Minjoon Seo and Hannaneh Hajishirzi and Ali Farhadi , title =. 2016 , booktitle = on Pith
  74. [102] Lei Li and Yuancheng Wei and Zhihui Xie and Xuqing Yang and Yifan Song and Peiyi Wang and Chenxin An and Tianyu Liu and Sujian Li and Bill Yuchen Lin and others , title =. 2025 , booktitle = on Pith
  75. [103] Niantong Li and Guangzheng Hu and Weixu Qiao and Ying Ba and Qichen Hong and Shijun Shen and Jinlin Wang and Fan Zhou and Jianye Kang and Xin Shang and others , title =. 2026 , journal = on Pith
  76. [104] Yaron Lipman and Ricky T. Q. Chen and Heli Ben-Hamu and Maximilian Nickel and Matt Le , title =. 2023 , booktitle = on Pith
  77. [105] Dongyang Liu and Peng Gao and David Liu and Ruoyi Du and Zhen Li and Qilong Wu and Xin Jin and Sihan Cao and Shifeng Zhang and Hongsheng Li and Steven Hoi , title =. 2026 , booktitle = on Pith
  78. [106] Jingyuan Liu and Jianlin Su and Xingcheng Yao and Zhejun Jiang and Guokun Lai and Yulun Du and Yidao Qin and Weixin Xu and Enzhe Lu and Junjie Yan and others , title =. 2025 , journal = on Pith
  79. [107] Shiyu Liu and Yucheng Han and Peng Xing and Fukun Yin and Rui Wang and Wei Cheng and Jiaqi Liao and Yingming Wang and Honghao Fu and Chunrui Han and others , title =. 2025 , journal = on Pith
  80. [108] Yuan Liu and Haodong Duan and Yuanhan Zhang and Bo Li and Songyang Zhang and Wangbo Zhao and Yike Yuan and Jiaqi Wang and Conghui He and Ziwei Liu and others , title =. 2024 , booktitle = on Pith

The first 100 of 178 references are listed here. All of them are in this paper's citation record.

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment