DynThinker: Unifying Static 3D and Dynamic 4D Spatial Reasoning

Qihang Cao; Haoyuan Li; Zihan Guo; Sensen Gao; Zihui Gao; Huixin Zhang; Fan He; Jianhua Han

cs.CVOct 14, 2026

DynThinker: Unifying Static 3D and Dynamic 4D Spatial Reasoning

Authors: Qihang Cao, Haoyuan Li, Zihan Guo, Sensen Gao, Zihui Gao, Huixin Zhang, Fan He, Jianhua Han,

Organizations: 1Nanyang Technological University · 2Shenzhen campus of Sun Yat-sen University · 3Yinwang Intelligent Technology Co. Ltd · 4MBZUAI

Abstract

Achieving general spatial intelligence requires a unified understanding of both static 3D and dynamic 4D. However, we find that existing models struggle to excel at both: 3D post-training can weaken dynamic understanding, while 4D-specialized models often lack strong static 3D reasoning. Our analysis reveals that this spatial--dynamic interference is highly non-uniform across parameter groups, with the balance between static and dynamic capabilities particularly sensitive to changes in late-layer MLPs. To address this interference, we propose DynThinker, a unified framework for jointly strengthening static and dynamic spatial reasoning. Decoupled Exponential Moving Average (D-EMA) applies configurable group-wise averaging to constrain drift in interference-sensitive pretrained parameters while preserving flexibility for capability enhancement. Spatial-Dynamic Fusion (SDF) complements D-EMA by selectively integrating spatial and temporal priors from a dynamic geometry encoder into intermediate VLM representations under semantic guidance. Across ten metrics spanning nine static 3D and dynamic 4D benchmarks, DynThinker achieves the highest average score among all evaluated models. Notably, DynThinker achieves 56.0 on ReVSI, 67.2 on DSR-Bench, and 51.2 and 40.1 under the sample-wise and group-wise evaluation protocols of DSI-Bench, respectively, setting new state-of-the-art results among the compared open-source models across all four evaluations within a unified model.

Figures & tables

Figure 2: Spatial–dynamic interference and capability preservation. (A) Interpolation between Qwen3-VL-8B and its 3D-specialized counterpart ( Li et al., 2026 ) improves VSI-Bench but degrades VLM4D, revealing a clear spatial–dynamic interference. (B) Restoring selected parameter groups toward their pretrained weights reveals that dynamic performance is particularly sensitive to post-training changes beyond the first four MLPs ( $j>4$ ). (C) D-EMA uses configurable temporal averaging to consolidate interference-sensitive pretrained parameters during adaptation. As MLLM4D ( Yin et al., 2026 ) training data scales up, D-EMA consistently improves both static 3D and dynamic 4D performance, whereas training without D-EMA leads to degradation in both.

Figure 3: Overview of the DynThinker architecture. Our framework integrates complementary spatial and dynamic priors from a dynamic geometry encoder through Spatial-Dynamic Fusion (SDF) layers. Importance Gating uses the VLM’s semantic context to selectively modulate both feature streams, enabling joint reasoning about geometric structure and temporal dynamics.

Models	AVG.	Spatial statics						Temporal dynamics
Models	AVG.	VSI	MMSI	MindCube	ViewSpatial	SITE	CV-Bench	VLM4D	DSR	DSI-S	DSI-G
Human	-	79.2	97.2	94.5	-	67.5	-	98.8	-	-	-
Random Choice	-	34.0	25.0	33.0	26.3	0.0	-	24.1	-	25.0	0.1
Proprietary Models
Seed-1.6 ( ByteDance Seed, 2025 )	-	49.9	38.3	48.7	43.8	54.6	85.2	-	-	41.3	16.1
Gemini-2.5-Pro ( Comanici et al., 2025 )	50.6	53.5	38.0	57.6	46.0	57.0	85.9	62.0	31.7	46.9	27.1

Table 1: Cross-benchmark comparison. AVG is the unweighted mean across all ten metrics across nine benchmarks and is reported only for models with complete coverage. Bold marks our column-best results among the listed open-source models.

\dagger

indicates evaluation without token compression. * indicates trained with S1+S2 dataset setting from VG-LLM. Benchmarks include VSI-Bench ( Yang et al., 2025a ) , MMSI-Bench ( Yang et al., 2025c ) , MindCube ( Yin et al., 2025 ) , ViewSpatial ( Li et al., 2025a ) , SITE ( Wang et al., 2025c ) , CV-Bench ( Tong et al., 2024 ) , VLM4D ( Zhou et al., 2025b ) , DSR ( Zhou et al., 2025a ) , DSI ( Zhang et al., 2025b ) (S: sample-wise, G: group-wise).

Methods	Frames	Avg.	Numerical Answer				Multiple-Choice Answer
Methods	Frames	Avg.	Obj.Count	Abs. Dist.	Obj. Size	Room Size	Rel. Dist.	Rel. Dir.	Route Plan
Proprietary Models
GPT-5.2 ( OpenAI, 2025b )	64	50.9	56.2	41.5	73.9	63.0	48.4	34.9	38.2
Gemini 3 Flash ( Google DeepMind, 2025 )	1 FPS	57.6	65.7	53.1	77.6	52.8	64.6	47.9	41.8
Gemini 3 Pro ( Gemini, 2025 )	1 FPS	60.9	60.1	54.7	79.3	51.9	68.1	56.0	56.4
Open-source Models

Table 2: Performance comparison on ReVSI-Bench. ReVSI adapts ground-truth answers to the evaluated frame count, assessing spatial reasoning under different frame budgets. DynThinker achieves the highest overall accuracy among the compared open-source models.

Model	Components					Results
	SFT	D-EMA	VDPM	Static	Tem.	VSI	VLM4D
Qwen3-VL-4B	✗	✗	✗	✗	✗	54.90	57.10
GeoThinker Qwen3-VL-4B	✓	✗	✗	✗	✗	51.31	45.20
DynThinker Qwen3-VL-4B	✓	✓	✗	✗	✗	53.14	53.57
	✓	✓	✓	✗	✗	53.60	54.01
	✓	✓	✓	✓	✗	56.19	58.64

Table 3: Component ablations and dynamic-data scaling. (a) From GeoThinker with the same Qwen3-VL-4B backbone, we progressively add D-EMA, VDPM, spatial and temporal priors, then scale the backbone to 8B. (b) MLLM4D data scaling on the VLM4D Real split, with reported MLLM-4D SFT results included for comparison. The dashed line marks the previous SOTA MLLM-4D RFT score of 63.1, while the lower panel separately reports overall VLM4D accuracy.

Appendix figures & tables18 assets

Supplementary material from the paper’s appendix.

Appendix

Model	Benchmark	# Frames
Model	Benchmark	16	32	64	128
GeoThinker Qwen3-VL-8B-32frame	VSI	69.2	72.6	73.4	73.4
	VSI-Debiased	64.3	66.3	67.7	68.1
	ReVSI	52.6	53.9	54.8	55.9
DynThinker Qwen3-VL-8B-32frame	VSI	64.0	69.3	70.7	72.1
	VSI-Debiased	57.1	63.1	64.4	66.5

Appendix

Table 4: Comparisons of zero-shot frame scaling. Both models use Qwen3-VL-8B and are trained with 32-frame inputs. We vary the evaluation frame budget from 16 to 128 on VSI-Bench, VSI-Debiased, and ReVSI-Bench. ReVSI-Bench adjusts ground-truth answers according to the evaluated frame count.

Training Configuration	Target D-EMA Drift	Avg.	Spatial Statics						Temporal Dynamics
Training Configuration	Target D-EMA Drift	Avg.	VSI	MMSI	MindCube	ViewSpatial	SITE	CV-Bench	VLM4D	DSR	DSI-S	DSI-G
Base mixture, 8-frame training	12.5%	50.28	61.5	31.9	42.7	42.9	57.1	86.6	62.9	46.6	45.2	25.4
Base mixture, 8-frame training	25%	54.77	64.8	30.7	67.1	46.7	56.8	87.4	61.8	61.2	45.2	26.0
Base mixture, 32-frame training	25%	53.73	66.2	30.4	61.2	45.6	58.0	87.0	61.3	59.5	44.3	23.9
Base mixture, 32-frame training	50%	57.24	69.2	29.1	87.3	47.4	56.7	87.4	57.0	66.1	45.2	27.0
+ MLLM4D: $10\%\rightarrow 50\%$	50%	59.87	70.1	31.4	90.2	48.0	57.0	86.3	58.6	67.7	50.5	39.0

Appendix

Table 5: Ablations of the scaled training configuration. All rows use DynThinker with Qwen3-VL-8B. The base mixture follows GeoThinker’s data recipe without PhysGame ( Cao et al., 2024 ) or the additional 430K general video samples, and includes 10% of MLLM4D. The final two rows cumulatively expand this mixture. Target drift budget denotes the target drift budget for pretrained parameters.

Models	Rank	Avg.	Static Understanding			Dynamic Understanding
			Dim. Meas.	Spatial Relation	3D Video Grounding	Disp. & P.L.	Speed & Acc.	Ego Orient.	Traj. Desc.	Pose Est.
Proprietary Models (API)
GPT-4o	10	34.8	27.1	51.8	29.0	23.2	35.4	33.7	32.0	53.6
Gemini-2.0-Flash	6	38.7	31.9	50.0	31.8	27.7	32.1	10.8	38.5	61.3
Claude-3.7-Sonnet	3	40.5	29.8	45.5	35.7	28.9	38.8	40.0	47.4	62.6
Gemini-2.5-Pro	1	41.4	38.7	53.8	36.9	33.9	33.1	52.5	47.4	50.4

Appendix

Table 6: Evaluation on STI-Bench. Orange = best and Light Orange = second best. Note: The random baseline for all tasks is 20% and @30f denotes evaluated with 30 frames.

Figure 4: Comparison between GeoThinker and DynThinker across static spatial, dynamic, video understanding, and embodied benchmarks. On static spatial tasks (left), DynThinker improves on several benchmarks while obtaining competitive scores on VSI-Bench, SPAR, and VSTI. On dynamic, video, and embodied tasks (right), DynThinker improves broadly, with MVBench as the main exception.

Figure 5: Physical false-premise rejection on real videos. DynThinker correctly identifies absent objects, preserving the pretrained model’s grounded responses where spatially specialized models hallucinate motion directions.

Figure 6: Physical false-premise rejection on synthetic videos. DynThinker correctly rejects questions about nonexistent griffins and unicorns, while VG-LLM and GeoThinker assign unsupported motion directions.

Figure 7: Motion and rotation understanding on real VLM4D videos. DynThinker correctly answers stationary-object and perspective-dependent rotation questions that are missed by both the pretrained and spatially specialized models.

Figure 8: Motion understanding on synthetic VLM4D videos. DynThinker preserves correct pretrained predictions for lateral motion and motion away from the camera, where spatially specialized models predict opposite directions.

Model	Real(%)	Synthetic(%)	Format(%)		Overall
Model	Real(%)	Synthetic(%)	Empty	Valid	Overall
MLLM4D Official
Qwen3-VL-8B	52.1	52.4	–	–	52.2
MLLM4D Qwen2.5-VL-7B	59.4	49.7	–	–	57.0
MLLM4D Qwen3-VL-8B	63.1	54.4	–	–	61.0
MLLM4D-SFT	59.7	–	–	–	–

Appendix

Table 7: Comparison of official and deterministic evaluation protocols on VLM4D. The upper block reproduces results reported by MLLM4D, whose official evaluation uses an auxiliary language model to extract the predicted option from model-generated responses. The lower block reports results obtained with our deterministic protocol, which directly matches the output against the valid options A–D without an auxiliary evaluator. To isolate answer parsing from differences in inference frameworks or evaluation prompts, rows marked with

*

are rescored directly from the model responses released in the official MLLM4D repository rather than regenerated using our framework. Empty denotes responses containing no predicted answer, while Valid denotes outputs that directly match one of the four option labels. These two columns evaluate output format rather than answer correctness. Real and synthetic averages, as well as the overall score, are weighted by their corresponding sample counts.

\dagger

denotes the final DynThinker checkpoint evaluated without token compression, and

\ddagger

denotes the best DynThinker checkpoint trained under the MLLM4D data-scaling setting. Dashes indicate unavailable results.

Methods	Avg.	Numerical Answer				Multiple-Choice Answer
Methods	Avg.	Obj.Count	Abs. Dist.	Obj. Size	Room Size	Rel. Dist.	Rel. Dir.	Route Plan	Appr. Order
Baseline
Chance Level (Random)	–	–	–	–	–	25.0	36.1	28.3	25.0
Chance Level (Frequency)	34.0	62.1	32.0	29.9	33.1	25.1	47.9	28.4	25.2
Proprietary Models (API)
GPT-4o ( Hurst et al., 2024 )	34.0	46.2	5.3	43.8	38.2	37.0	41.3	31.5	28.5

Appendix

Table 8: Performance comparisons on VSI-Bench (Vanilla regime). DynThinker matches or outperforms GeoThinker and other baseline across different model scales, achieving the best overall performance with the Qwen3-VL-8B variant.

Strategy	Target Drift Budget		VSI-Bench	VLM4D
Strategy	MLP ( $j>4$ )	Others	VSI-Bench	VLM4D
Qwen2.5-VL-7B
Pretrain	–	–	29.34	43.80
D-EMA	12.5%	12.5%	34.44 (+5.10)	55.17 (+11.37)
	100%	100%	49.65 (+20.31)	46.64 (+2.84)
	12.5%	100%	50.37 (+21.03)	48.56 (+4.76)

Appendix

Table 9: Ablation of D-EMA drift allocation. All post-trained models are trained under the spatial-reasoning vanilla regime with SDF enabled. We vary the target drift budgets for MLP layers

j>4

and other pretrained parameters. Newly introduced SDF parameters are excluded from averaging. (

\cdot

) denotes the performance change compared to pre-train model.

Training Configuration	VSI-Bench	VLM4D
Qwen2.5-VL-7B
Pretrained	29.34	43.80
Static 3D SFT	49.37	46.80
Static 3D SFT + D-EMA	49.73	47.57
Static 3D SFT + D-EMA + SDF	50.37	48.56
Qwen3-VL-4B

Appendix

Table 10: Training and component ablations under the Static 3D-reasoning vanilla regime. We compare pretrained backbones, Static 3D SFT, EMA-based preservation, and SDF integration. For Qwen2.5-VL-7B, D-EMA uses target drift budgets of 12.5% for MLP layers

j>4

and 100% for other pretrained parameters. For Qwen3-VL, D-EMA uses a shared budget of 12.5%.

Strategy	Update interval	VSI-Bench	VLM4D
Baseline
Pre-train Model	–	54.90	57.10
Direct SFT	–	52.61	46.75
Frozen MLP
Frozen All MLP	–	51.41	50.27
Weight Pulling

Appendix

Table 11: Comparison of capability-preservation strategies under the temporal-dynamic vanilla regime. We compare MLP freezing, weight pulling, spectral regularization, conventional EMA, and D-EMA using the same training mixture and 8-frame training inputs. All results use Qwen3-VL-4B unless otherwise specified. Interval step denotes the number of optimization steps between applications of the corresponding strategy.

Model	Benchmark	# Frames
Model	Benchmark	8	16	32	64	128
With token compression
DynThinker Qwen3-VL-4B	VSI	47.72	49.37	53.14	56.37	57.24
DynThinker Qwen3-VL-4B	VLM4D	57.65	53.41	53.41	53.02	51.87
DynThinker Qwen3-VL-8B	VSI	49.93	53.52	57.11	58.87	OOM
DynThinker Qwen3-VL-8B	VLM4D	62.33	61.56	60.51	60.51	OOM

Appendix

Table 12: Inference-time frame ablation under the temporal-dynamic vanilla regime. All models are trained with 8-frame inputs and evaluated with different frame budgets on VSI-Bench and VLM4D, with and without visual token compression.

\ast

denotes sparse VDPM decoding with a maximum of 8 sampled frames, as described in Model Configurations. unmarked variants decode all input frames. OOM indicates an out-of-memory failure.

Figure 9: Layer-boundary analysis for selective MLP preservation. We index the LLM blocks from $1$ to $L$ , and let $b$ denote the number of earliest MLP blocks that retain their specialized weights. Starting from GeoThinker-Qwen3-VL-8B, we restore MLP weights in layers $j>b$ to their pretrained Qwen3-VL-8B values, while retaining the specialized weights in layers $j\leq b$ and all other parameters. Thus, $b=0$ restores all MLP layers, whereas $b=4$ retains the first four specialized MLP layers. We report VLM4D and VSI-Bench performance as the boundary varies. The dashed line marks our selected boundary $b=4$ .

Model	Interpolate	VSI	VLM4D
Model	ratio	VSI	VLM4D
Baseline
GeoThinker Qwen3-VL-8B ( $\theta_{T}$ )	100%	72.6	42.40
Qwen3-VL-8B-Instruct ( $\theta_{0}$ )	0%	57.75	60.07 (60.07)
Interpolate all
75% $\theta_{T}$ + 25% $\theta_{0}$	75%	71.70	45.95

Appendix

Table 13: Parameter-group interpolation between GeoThinker and the pretrained Qwen3-VL-8B model. We interpolate the specialized checkpoint

\theta_{T}

with the pretrained checkpoint

\theta_{0}

, either globally or within selected parameter groups. The interpolation ratio denotes the contribution of

\theta_{T}

in the selected group. Remaining parameters retain their specialized values. Early and late MLP groups correspond to layer indices

j\leq 4

and

j>4

, respectively. Scores outside parentheses use deterministic rule-based answer parsing, while parenthesized values report direct exact-match accuracy without parsing.

Model	Interpolate	VSI	VLM4D
Model	ratio	VSI	VLM4D
Baseline
DynThinker Qwen3-VL-4B w/o D-EMA ( $\theta_{T}$ )	100%	71.06	50.33
Qwen3-VL-4B-Instruct ( $\theta_{0}$ )	0%	54.90	57.10 (1.87)
Interpolate all
75% $\theta_{T}$ + 25% $\theta_{0}$	75%	71.32	48.84

Appendix

Table 14: Parameter-group interpolation for DynThinker without D-EMA. We interpolate the spatially specialized Qwen3-VL-4B checkpoint

\theta_{T}

with its pretrained counterpart

\theta_{0}

, globally or within selected parameter groups. The interpolation ratio denotes the contribution of

\theta_{T}

in the interpolated group. All remaining parameters retain their specialized values. Early and late MLP groups correspond to layer indices

j\leq 4

and

j>4

, respectively. Scores outside parentheses use deterministic rule-based answer parsing, while parenthesized values report direct exact-match accuracy without parsing.

Model	Steps	Spatial	Object	Goal	Long	Avg
Baseline
$\pi_{0}$ +FAST ( Pertsch et al., 2025 )	-	96.4	96.8	88.6	60.2	85.5
$\pi_{0}$ ( Black et al., 2024 )	-	96.8	98.8	95.8	85.2	94.1
OpenVLA-OFT-7B ( Kim et al., 2025 )	175K	97.6	98.4	97.9	94.5	97.1
GR00T-N1.5 ( Bjorck et al., 2025 )	20K	92.0	92.0	86.0	76.0	86.5
Qwen3-VL-4B

Appendix

Table 15: Comparison of different VLA models on LIBERO. We train one policy for all 4 suites. All scores are averaged over 500 trials for each task suite (10 tasks × 50 episodes).

Spatiotemporal Reasoning

Spatiotemporal reasoning studies how systems connect objects, events, locations, and changes across space and time.

See all 70 papers

4DThinker: Thinking with 4D Imagery for Dynamic Spatial Understanding
May 7, 2026Zhangquan Chen, Manyuan Zhang, Xinlei Yu +9Visual Spatial Reasoning Video Reasoning
Tsinghua University, SIGS · Meituan · The Chinese University of Hong Kong +3
Thinking in Dynamics: How Multimodal Large Language Models Perceive, Track, and Reason Dynamics in Physical 4D World
Mar 13, 2026Yuzhi Huang, Kairun Wen, Rongxin Gao +14Spatiotemporal Reasoning Multimodal Large Language Models
XMU · SUSTech · THU +4
DynTrace: Tracking Dynamic Object Evidence for 4D Spatio-Temporal Reasoning in MLLMs
Jul 14, 2026Rongxin Gao, Yuzhi Huang, Dongxuan Liu +8Spatiotemporal Reasoning Visual Object Tracking
1Xiamen University · 2Shenzhen International Graduate School, Tsinghua University · 3China University of Mining and Technology +1

May 7, 2026J/K move · Enter open · S save

4DThinker: Thinking with 4D Imagery for Dynamic Spatial Understanding

Zhangquan Chen, Manyuan Zhang, Xinlei Yu, Xiang An+8

Tsinghua University, SIGS · Meituan · The Chinese University of Hong Kong+3

Visual Spatial Reasoning Video Reasoning VLM Reasoning 4D Scene Understanding Spatial Reasoning

Open paper AskPDF arXiv

Dynamic spatial reasoning from monocular video is essential for bridging visual intelligence and the physical world, yet remains challenging for vision-language models (VLMs). Prior approaches either verbalize spatial-temporal reasoning entirely as text, which is inherently verbose and imprecise for complex dynamics, or rely on external geometric modules that increase inference complexity without fostering intrinsic model capability. In this paper, we present 4DThinker, the first framework that enables VLMs to "think with 4D" through dynamic latent mental imagery, i.e., internally simulating how scenes evolve within the continuous hidden space. Specifically, we first introduce a scalable, annotation-free data generation pipeline that synthesizes 4D reasoning data from raw videos. We then propose Dynamic-Imagery Fine-Tuning (DIFT), which jointly supervises textual tokens and 4D latents to ground the model in dynamic visual semantics. Building on this, 4D Reinforcement Learning (4DRL) further tackles complex reasoning tasks via outcome-based rewards, restricting policy gradients to text tokens to ensure stable optimization. Extensive experiments across multiple dynamic spatial reasoning benchmarks demonstrate that 4DThinker consistently outperforms strong baselines and offers a new perspective toward 4D reasoning in VLMs. Our code is available at https://github.com/zhangquanchen/4DThinker.

Abstract

Spatiotemporal Reasoning

Explore similar work

4DThinker: Thinking with 4D Imagery for Dynamic Spatial Understanding

Thinking in Dynamics: How Multimodal Large Language Models Perceive, Track, and Reason Dynamics in Physical 4D World

DynTrace: Tracking Dynamic Object Evidence for 4D Spatio-Temporal Reasoning in MLLMs