On-policy distillation (OPD) has emerged as an effective post-training approach to improving fine-grained perception in multimodal large language models by amortizing costly inference-time zooming operations into a single-pass recipe. However, vanilla OPD is inherently susceptible to linguistic momentum: the teacher can follow the student's erroneous prefixes, thus weakening corrective supervision. To address this, we propose Blink On-Policy Distillation (BlinkOPD), in which the teacher "blinks" at every token: it reads the same prefix once with its eyes open and once with its eyes closed. A first-glance gain further strengthens the correction on rollouts that conflict with the teacher. By extracting more effective supervision from each on-policy trajectory, BlinkOPD achieves state-of-the-art average performance across six fine-grained benchmarks using only 500 training images (12x less data) than existing OPD methods. Beyond fine-grained perception, BlinkOPD also demonstrates strong zero-shot improvements across ten held-out benchmarks spanning general visual perception, visual hallucination, and multimodal reasoning abilities. Code: https://github.com/Supltz/BlinkOPD
Figures & tables
Figure 1: Comparison of existing OPSD methods on four visual abilities across 16 benchmarks using Qwen3.5-4B as the base model. With 12 × fewer training images, BlinkOPD outperforms competing methods on four visual abilities across 16 benchmarks.
Figure 2: Comparison of the distillation targets of Vision-OPD, VAD, CV-OPD, OPD-V, OPD-Aha and our proposed BlinkOPD on linguistic momentum and privileged evidence survival. (a) As the teacher reads more of the wrong prefix 1 1 1 We also provide a control experiment that replaces the prefix with another question’s answer of the same length, which confirms that this rise comes from the content of the prefix rather than its length ( Appendix B ). , the follow rate tends to rise: Vision-OPD rises from 23% to 100%, VAD, CV-OPD and OPD-V from about 50% to 100%, and OPD-Aha still ends at 64%, whereas BlinkOPD stays below 40% throughout. (b) As the teacher reads more of the wrong prefix, the survival score tends to decrease: Vision-OPD, VAD, CV-OPD and OPD-V turn negative within the first quarter of the prefix and never recover, and OPD-Aha also falls below zero by the end, whereas BlinkOPD remains positive and outperforms the others throughout.
Figure 3: Comparison of contrastive references on residual drift and false belief. (a) Image-based references leave 20–23% of the drift uncancelled, whereas the text-only and eyes-closed references leave about 10%. (b) All other references raise the probability of “not present” false belief to 0.39–0.63. Only our proposed eyes-closed reference is low on both measures.
Figure 4: Contrastive reference construction for OPSD. (a) Image-based references replace the crop with gray, black, noisy, or mismatched images [ 23 , 16 , 2 , 14 , 30 ] . (b) The text-only reference removes the image from the conversation [ 35 ] . (c) Our eyes-closed reference removes only the image-content tokens while preserving vision markers and original positional indices.
Figure 5: Overview of BlinkOPD. The student generates rollouts from the full image I and question q , together denoted by xs . The frozen privileged teacher θ0 reads each student prefix twice: once with its eyes open on the zoomed crop ( pt+ ), and once with its eyes closed ( pt∅ , Sec. 3.1 ). Their log-ratio rt keeps the privileged visual evidence, while the prefix-induced linguistic drift shared by both cancels. A prefix-free first glance of the same teacher flags rollouts whose answer conflicts with it, and the gain is raised by a factor γ on these rollouts ( Sec. 3.2 ). The student is trained toward the resulting distillation target ( Sec. 3.3 ). At test time, the student runs alone in a single pass.
Fine-Grained
High-Resolution
Real-World
Method
Size
Teacher
#Img
V ∗
ZoomBench
HR-4K
HR-8K
MME-RW EN
MME-RW CN
Avg.
Δ
Proprietary models (reported by [ 36 , 12 ] ; also evaluated in [ 37 , 30 , 23 ] )
GPT-5.2 [ 19 ]
–
–
–
79.1
50.9
81.1
78.4
72.6
68.8
71.8
GPT-5.4 [ 21 ]
–
–
–
77.0
52.7
84.0
77.9
74.2
70.9
72.8
Gemini 3.1 Pro [ 9 ]
–
–
–
88.0
61.2
89.6
86.9
76.5
73.3
79.3
Gemini 3.5 Flash [ 13 ]
–
–
–
89.0
61.4
89.1
86.6
75.3
74.0
79.2
Table 1: Fine-grained perception. Accuracy (%) on six benchmarks. Gray : results reported by the original papers or reproduced by others; listed for reference only. Agentic models and ZwZ are built on older backbones ( ‡ Qwen2.5-VL-7B, § Qwen3-VL-8B). Bold and underline mark the best and second best within each re-evaluated block. Teacher: Self is a privileged self-teacher (a copy of the student on the zoomed crop); V-Zero instead distills from a Qwen3.5-27B teacher. #Img: unique training images. Δ : gain over the base model.
General Perception
Hallucination
Multimodal Reasoning
Average
Method
MMStar
CV-Bench
MMVP
POPE
HallB.
MathVista
MathVerse
MATH-V
We-Math
DynaMath
Perc.
Hall.
Reas.
All
Δ
Qwen3.5-4B
Base model [ 24 ]
68.3
87.0
58.7
89.5
76.1
82.0
74.2
63.9
58.8
57.0
71.3
82.8
67.2
71.0
–
Vision-OPD [ 36 ]
71.5
85.6
59.3
89.4
71.7
81.2
71.8
60.9
63.3
63.9
72.1
80.6
68.2
73.1
+2.1
OPD-V [ 2 ]
71.4
85.8
58.7
89.5
73.3
80.5
72.3
61.6
64.9
61.1
72.0
81.4
68.1
73.2
+2.2
CV-OPD [ 30 ]
71.5
86.6
56.7
89.3
71.8
81.0
72.4
63.5
67.1
57.8
71.6
80.6
68.4
73.4
+2.4
Table 2: Zero-shot transfer to ten held-out benchmarks. Perc., Hall. and Reas.: average accuracy on general perception, hallucination mitigation and multimodal reasoning. All: average over all 16 benchmarks, including the six fine-grained perception benchmarks of Tab. 1 . Δ : change of All relative to the base model. Bold and underline : best and second best within each model size.
Figure 6: Corrective supervision and data efficiency with Qwen3.5-4B. (a) As the teacher reads more of the wrong prefix, the corrective density tends to decrease: Vision-OPD drops from 68% to 0%, VAD, CV-OPD and OPD-V from about 40% to 0%, and OPD-Aha declines to 28%, whereas BlinkOPD stays above 40% throughout. (b) Mean accuracy across the six fine-grained benchmarks at different training scales. BlinkOPD leads at every scale and reaches 79.6% with 6.2K images.
Variant with Qwen3.5-4B as the base model
Avg.
Δ
BlinkOPD (0.5K)
79.1
–
Ablation
w/o first-glance gain (same β for all rollouts)
78.5
−0.6
→ gray-image reference
77.4
−1.7
→ black-image reference
75.4
−3.7
→ Gaussian-noise-image reference
75.6
−3.5
Table 3: Ablation&sensitivity analysis of BlinkOPD (0.5K) on 4B.
Figure 7: Follow rate of the raw teacher (Vision-OPD) on the 1,732 student-wrong rollouts when it reads the student’s own prefix or a prefix of another question with the same length. The thin gray line marks the follow rate before reading any prefix (23.2%).
Figure 8: Residual drift δ (%, Eq. 5 ) when the teacher and the reference read (a) 50%, (b) 75% or (c) all of the student prefix ( Fig. 3 a).
Figure 9: Trade-off of a global gain on the probe of Sec. 2.2 . Each point is the share of rollouts on which the target picks the correct option, averaged over the read-out points τ=25 , 50, 75 and 100%. Raising the global gain β raises the corrective density C on student-wrong rollouts but lowers the retention R on student-correct ones.
Figure 10: First-glance gain ( γ=2 ) against the global gain on the probe of Sec. 2.2 : C against R , each averaged over the four read-out points, as β is swept from 0 to 12 for the global gain and from 0 to 6 for the first-glance gain, so that both reach a largest gain of 12. The first-glance gain lies above the global gain.
Method
Teacher
Reference
Role of the reference
Gain
Granularity
Labels
Vision-OPD [ 36 ]
EMA
none
–
–
–
no
OPD-V [ 2 ]
EMA
crop with a random black patch
on/off weight on each token’s loss
–
per token
no
CV-OPD [ 30 ]
EMA
another region of the same image
hinge term that keeps the student closer to the teacher
fixed weight
per token
yes ∗
VAD [ 37 ]
EMA
pixelated crop
projects the teacher–student shift onto the evidence direction
Table 7: Comparison with the reproduced OPSD baselines. Granularity: the level at which the strength of the reference varies. Labels: whether ground-truth answers are used during training ( ∗ CV-OPD skips the questions on which no rollout is correct).
Figure 11: Qualitative examples of fine-grained perception. Left: the test image; the red box marks the region that the question refers to and is drawn only for this figure, since all models see the original image. Middle: the boxed region, enlarged. Right: the question, its options and the ground truth. Bottom: the answer of each model ( ✓ correct, ✗ wrong).
Figure 12: Qualitative examples of general perception. Left: the test image. Right: the question, its options and the ground truth. Bottom: the answer of each model ( ✓ correct, ✗ wrong).
Figure 13: Qualitative examples of hallucination. The layout is the same as in Fig. 12 .
Figure 14: Qualitative examples of multimodal reasoning. The layout is the same as in Fig. 12 .
Multimodal Large Language Models (MLLMs) still struggle with fine-grained visual understanding, where answers often depend on small but decisive evidence in the full image. We observe a regional-to-global perception gap: the same MLLM answers fine-grained questions more accurately when conditioned on evidence-centered crops than on the corresponding full images, suggesting that many failures stem from difficulty to focus on relevant evidence rather than insufficient local recognition ability. Motivated by this observation, we propose Vision-OPD (Vision On-Policy Distillation), a regional-to-global self-distillation framework that transfers the model's own privileged regional perception to its full-image policy. Vision-OPD instantiates two conditional policies from the same MLLM: a crop-conditioned teacher and a full-image-conditioned student. The student generates on-policy rollouts, and Vision-OPD minimizes token-level divergence between the teacher and student next-token distributions along these rollouts. This enables the model to internalize the benefit of visual zooming without external teacher models, ground-truth labels, reward verifiers, or inference-time tool use. Experiments on multiple fine-grained visual understanding benchmarks show that Vision-OPD models achieve competitive or superior performance against much larger open-source, closed-source, and "Thinking-with-Images" agentic models. The code is available at https://github.com/VisionOPD/Vision-OPD
Qianhao Yuan, Jie Lou, Xing Yu +4
Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences · University of Chinese Academy of Sciences · Xiaohongshu Inc.
On-policy distillation (OPD) improves reasoning by providing token-level supervision from a teacher on a student's own trajectories. Existing methods primarily focus on enhancing this teacher-side guidance (e.g., by enriching teacher inputs and refining teacher feedback), yet we find that limited student perception is another critical bottleneck in multimodal OPD. By providing oracle visual facts, the performance of OPD-trained students can still be substantially improved for both weak and strong teachers. To address this bottleneck, we propose S-OPD, a simple multimodal on-policy distillation framework that explicitly strengthens student perceptual learning through two objectives. Specifically, Teacher-calibrated Policy Contrast separates student policies under original and masked images with teacher-based token-level gating, strengthening the student's reliance on visual evidence during reasoning. Policy Agreement aligns student policies under original and noise-perturbed images, further improving perceptual robustness to visual noise. Notably, our method can be seamlessly plugged into existing OPD frameworks, requiring no additional data annotations, model parameters or inference operations. Extensive experiments on eight benchmarks across student scales and distillation paradigms demonstrate consistent performance improvements, with gains of up to 4.25 points on LogicVista. When combined with existing teacher-side supervision methods, our method can yield further gains. Code is available at https://github.com/Sirilaw/S-OPD.
Siyuan Liu, Kanghui Tian, Yue Duan +4
Nanjing University · Fudan University · Nanjing University of Posts and Telecommunications
Multimodal large language models (MLLMs) often struggle with fine-grained visual perception when processing complete images, as critical evidence may only appear in local regions. On-policy self-distillation (OPD) enables transferring privileged visual knowledge from informative views to full-image policies, but querying the teacher for every rollout introduces substantial supervision costs. In this work, we propose BAS-OPD, a budget-aware selective OPD framework that allocates teacher supervision under limited query budgets. Instead of querying all rollouts, BAS-OPD selects informative samples while maintaining full-batch student generation. We explore random, uncertainty-based, and learned utility-based selection strategies, where the learned selector estimates query value from detached rollout statistics and online utility signals derived from student--teacher agreement and teacher confidence without additional student forward passes. BAS-OPD only changes training-time supervision allocation and preserves single-pass full-image inference. Experiments on fine-grained multimodal perception benchmarks demonstrate that BAS-OPD achieves strong performance while substantially reducing teacher supervision costs, highlighting the effectiveness of selective OPD under constrained budgets.
Zihan Chen, Hengguang Zhou, Yuan Kang +5
HFIPS, Chinese Academy of Sciences · University of Science and Technology of China · University of California, Los Angeles