VisPhyWorld: Code-Based Physical Reasoning

• VisPhyWorld is an execution-based framework that converts visual observations into executable simulator code, making physical hypotheses inspectable, editable, and falsifiable.
• It evaluates models over 209 scenes from 108 physical templates by reconstructing videos under deterministic simulation, ensuring rigorous comparison against ground truth.
• The framework reveals that strong semantic scene understanding does not guarantee precise physical parameter inference, highlighting challenges in consistent motion simulation.

to=arxiv_search.search อาคารจีเอ็มเอ็ม json {"^^^^^^^^9query9 physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^^^^^", "^^^^^^^^9max_results9 ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9^^^^} to=arxiv_search.search 彩票天天乐json {"^^^^^^^^9query9 OR ^^^^^^^^9ti:\9 OR ^^^^^^^^9ti:\9 OR ^^^^^^^^9ti:\9 OR ^^^^^^^^9ti:\9 "^^^^^^^^9max_results9 ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9^^^^} VisPhyWorld is an execution-based framework for probing physical reasoning in multimodal LLMs by requiring them to generate executable simulator code from visual observations rather than merely answering recognition-style questions. The generated code is then sanitized, executed, and evaluated through reconstructed video, making the inferred world representation directly inspectable, editable, and falsifiable. In the formulation introduced with VisPhyBench, the framework operates on ^^^^9max_results9query99^^^^ evaluation scenes derived from ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9 OR ti:\9^^^^ physical templates, and its pipeline produces valid reconstructed videos in 9^^^^9 OR ti:\9^^^^.^^^^9 OR ti:\9^^^^% on the benchmark (&&&^^^^9query9^^^^&&&).

^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^. Conceptual basis

VisPhyWorld is motivated by a critique of dominant evaluation protocols for physical reasoning. Visual Question Answering and Violation of Expectation typically reward recognition and surface pattern matching, and can often be answered without committing to an explicit physical hypothesis. This is especially limiting for code-capable multimodal models, whose outputs are not predictive likelihoods over futures but structured artifacts such as scene descriptions, programs, or agentic plans. VisPhyWorld replaces answer selection with executable reconstruction: a model must transform visual observations into simulator code and reproduce scene dynamics under a deterministic execution contract.

This design yields three properties that the framework treats as central. First, the hypothesis is inspectable, because object states, forces, and parameters are explicit in code. Second, it is editable, because parameters such as friction or mass can be modified and the scene rerun. Third, it is falsifiable, because failure is observed by executing the program rather than inferred from linguistic inconsistency. This separates physical reasoning from rendering: the model is not primarily judged on whether it can describe a scene, but on whether it can instantiate a consistent world model that reproduces motion (&&&^^^^9query9^^^^&&&).

A recurrent misconception addressed by the framework is that strong semantic scene understanding is equivalent to strong physical reasoning. The reported experiments indicate the opposite pattern: state-of-the-art models often reconstruct object identity and coarse layout well, yet still struggle to infer physical parameters and to simulate consistent physical dynamics. Another misconception is that low motion error alone suffices; VisPhyWorld explicitly notes that RAFT-EPE must be interpreted jointly with holistic physics judgments because low EPE can arise from degenerate outputs.

^^^^9max_results9^^^^. Execution pipeline and world representation

The VisPhyWorld pipeline takes as input two key frames, denoted PRESERVED_PLACEHOLDER_^^^^9query9^^^^ and PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^, together with an optional detection context PRESERVED_PLACEHOLDER_^^^^9max_results9^^^^ containing object categories, bounding boxes, and colors in pixel space. An LLM agent then produces three intermediate artifacts: a motion analysis PRESERVED_PLACEHOLDER_^^^^9query9^^^^, a first-frame JSON scene specification PRESERVED_PLACEHOLDER_^^^^9ti:\9^^^^, and executable code PRESERVED_PLACEHOLDER_^^^^9 OR ti:\9^^^^. The code is validated, optionally repaired once using error-conditioned retry, and then executed deterministically to generate a reconstructed video PRESERVED_PLACEHOLDER_^^^^9 OR ti:\9^^^^, which is evaluated against ground truth PRESERVED_PLACEHOLDER_^^^^9 OR ti:\9^^^^ (&&&^^^^9query9^^^^&&&).

The framework prioritizes physics-capable backends. For the ^^^^9query9^^^^D subset it uses Three.js + Cannon.js, with a fixed perspective camera and white background. P^^^^9 OR ti:\9^^^^.js is also used as a code-generation target. For comparison and diagnosis, non-physics backends such as SVG and Manim are included; these often produce implausible contacts, including interpenetration. The ^^^^9max_results9^^^^D subset is generated with PHYRE. Coordinate conventions are explicit: the detection context uses image pixel coordinates with origin at the top left, the ^^^^9max_results9^^^^D subset uses a fixed orthographic camera, and the ^^^^9query9^^^^D subset uses a fixed perspective camera.

The generated program is expected to expose a structured world representation. Environment properties include gravity, time step, camera intrinsics, and background. Object definitions include primitive shapes such as spheres, boxes, lines or ramps, and U-shaped containers; geometry; material properties such as friction PRESERVED_PLACEHOLDER_^^^^9 OR ti:\9^^^^ and restitution $e$; and mass PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9^^^^. Initial states align positions and orientations to the first frame, with optional initial velocities or impulses used to match motion between PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^ and PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9max_results9^^^^. The simulator then steps the scene with a fixed time step and deterministic recording. The paper explicitly references rigid-body dynamics through PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9^^^^ and Coulomb-like friction through PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9ti:\9^^^^ (&&&^^^^9query9^^^^&&&).

This representation makes parameter-level diagnosis possible even though VisPhyBench v^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^ does not yet report parameter inference accuracy as a standalone metric. A plausible implication is that VisPhyWorld is less a benchmark of free-form video realism than a benchmark of whether a model can externalize a mechanically meaningful latent world model.

^^^^9query9^^^^. VisPhyBench benchmark design

VisPhyBench consists of ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9 OR ti:\9^^^^ physical templates instantiated into ^^^^9max_results9query99^^^^ videos with first-frame JSON annotations. The ^^^^9max_results9^^^^D subset is derived from and extends PhyWorld’s ^^^^9max_results9^^^^D data and is rendered via PHYRE-style scripts. The ^^^^9query9^^^^D subset is rendered with Three.js and simulated with Cannon.js. Scene types include stacks, ramps, collisions, slides, bounces, and topples, with objects such as spheres, boxes, planks, lines or ramps, and U-shaped containers (&&&^^^^9query9^^^^&&&).

The benchmark includes a full “sub” split of ^^^^9max_results9query99^^^^ scenes with difficulty labels assigned by eight annotators. Difficulty is partitioned as ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9ti:\9^^^^ easy, ^^^^9 OR ti:\9 OR ti:\9^^^^ medium, and ^^^^9max_results9 OR ti:\9^^^^ hard. A separate test split contains ^^^^9ti:\99^^^^ scenes, comprising ^^^^9max_results99^^^^ easy, ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9 OR ti:\9^^^^ medium, and ^^^^9query9^^^^ hard. Rendering is standardized through fixed camera settings, white background, deterministic time step, and fixed recording protocol.

For the ^^^^9max_results9^^^^D subset, object-type prevalence is reported directly: circle ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9query9^^^^.^^^^9query9^^^^% of scenes, line ^^^^9 OR ti:\9query9^^^^.^^^^9max_results9^^^^%, rectangle ^^^^9 OR ti:\9max_results9^^^^.^^^^9 OR ti:\9^^^^%, u-shape ^^^^9max_results9ti:\9^^^^.^^^^9 OR ti:\9^^^^%, triangle ^^^^9 OR ti:\9^^^^.^^^^9query9^^^^%, composite ^^^^9 OR ti:\9^^^^.^^^^9query9^^^^%. These statistics emphasize that the benchmark is not narrowly restricted to a single collision archetype but spans a controlled repertoire of rigid-body configurations.

The construction philosophy is programmatic. Templates are executable programs whose seeds sample placements, sizes, and sometimes material parameters. This preserves controllability while producing diversity across instantiations. The dataset is accompanied by prompt templates, validation and canonicalization procedures, a single-retry repair protocol, and a fallback template that guarantees well-defined evaluation. The project page is listed as https://tiger-ai-lab.github.io/VisPhyWorld/ (&&&^^^^9query9^^^^&&&).

^^^^9ti:\9^^^^. Evaluation protocol

VisPhyWorld uses a multi-metric protocol that evaluates reconstructed videos after temporal alignment. A reconstructed output counts as a valid reconstructed video if the generated code executes and produces a non-empty clip without resorting to fallback; one automatic repair attempt is allowed. At the pipeline level, the reported validity is 9^^^^9 OR ti:\9^^^^.^^^^9 OR ti:\9^^^^% (&&&^^^^9query9^^^^&&&).

The metric suite is deliberately heterogeneous. Appearance reconstruction is measured with PSNR and SSIM. Perceptual quality uses LPIPS, FSIM, VSI, and DISTS. Visual semantic consistency uses CLIP-Img and DINO cosine similarity. Text-video and analysis-text consistency uses CLIP-Cap, ROUGE-L, and BERTScore-F^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^. Motion and physical plausibility are assessed with RAFT-EPE, computed after coarse-to-fine temporal alignment using offset search and DTW. In addition, a Gemini-^^^^9max_results9^^^^.^^^^9 OR ti:\9^^^^-Pro holistic judge assigns a score from ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^ to ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9^^^^ and explicitly penalizes physical violations.

No single composite score is defined. This choice reflects the framework’s diagnostic orientation: different metrics isolate different parts of the failure surface. The paper notes, for example, that parameter inference accuracy is not reported in the current release, even though the code representation would make a metric of the form PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9 OR ti:\9^^^^ feasible in future work.

The evaluation philosophy differs from purely perceptual benchmarking in one important respect: a plausible video is not sufficient if it emerges from an opaque generator. VisPhyWorld evaluates the executable hypothesis itself. This makes failure attribution more precise, because errors can be localized to scene initialization, parameter choice, or constraint modeling rather than collapsed into a single visual score (&&&^^^^9query9^^^^&&&).

^^^^9 OR ti:\9^^^^. Empirical results and failure modes

The evaluated code-generation backbones are GPT-^^^^9 OR ti:\9^^^^, GPT-^^^^9ti:\9^^^^.^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^, Gemini-^^^^9query9^^^^-Pro, Claude Sonnet ^^^^9ti:\9^^^^.^^^^9 OR ti:\9^^^^, and Qwen^^^^9query9^^^^-VL-Plus, used with Three.js and P^^^^9 OR ti:\9^^^^.js. Pixel-space baselines are Stable Video Diffusion (SVD img^^^^9max_results9^^^^vid) and Veo-^^^^9query9^^^^.^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^. The central empirical finding is that models exhibit strong semantic scene understanding but difficulty with precise physical parameterization and consistent dynamics (&&&^^^^9query9^^^^&&&).

The reported results show a consistent backend effect. Three.js generally outperforms P^^^^9 OR ti:\9^^^^.js in reconstruction fidelity, and non-physics backends such as SVG and Manim reveal what happens when rigid-body solvers are absent: scripted motion, interpenetration, no response to impacts, or stationary objects when motion is expected. Among the Three.js systems, Gemini-^^^^9query9^^^^-Pro attains PSNR ^^^^9max_results9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^.^^^^9max_results9 OR ti:\9^^^^, SSIM ^^^^9query9^^^^.9^^^^9ti:\9ti:\9 OR ti:\9^^^^, success ^^^^9query9^^^^.9^^^^9 OR ti:\9 OR ti:\9^^^^, the lowest LPIPS ^^^^9query9^^^^.^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query999^^^^, and the highest CLIP-Img ^^^^9query9^^^^.^^^^9 OR ti:\99 OR ti:\9query9^^^^. GPT-^^^^9 OR ti:\9^^^^ (Three.js) reaches PSNR ^^^^9max_results9query9^^^^.^^^^9 OR ti:\9ti:\9^^^^, SSIM ^^^^9query9^^^^.9^^^^9query9 OR ti:\9query9^^^^, success ^^^^9query9^^^^.99^^^^9query9, DINO ^^^^9query9^^^^.^^^^9 OR ti:\9 OR ti:\9 OR ti:\9 OR ti:\9^^^^, and RAFT-EPE ^^^^9query9query9^^^^.^^^^9 OR ti:\9 OR ti:\9^^^^. The holistic Gemini score is highest for Gemini-^^^^9query9^^^^-Pro (Three.js) at ^^^^9query9^^^^.^^^^9 OR ti:\9query9^^^^, followed by GPT-^^^^9 OR ti:\9^^^^ (Three.js) at ^^^^9query9^^^^.^^^^9 OR ti:\9query9^^^^; the pixel-space baselines score lower on this axis, with Veo-^^^^9query9^^^^.^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^ at ^^^^9max_results9^^^^.^^^^9 OR ti:\9max_results9^^^^ and SVD at ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^.^^^^9ti:\9query9^^^^ (&&&^^^^9query9^^^^&&&).

System	Selected results	Observed significance
Gemini-^^^^9query9^^^^-Pro (Three.js)	PSNR ^^^^9max_results9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^.^^^^9max_results9 OR ti:\9^^^^, SSIM ^^^^9query9^^^^.9^^^^9ti:\9ti:\9 OR ti:\9^^^^, LPIPS ^^^^9query9^^^^.^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query999^^^^, holistic ^^^^9query9^^^^.^^^^9 OR ti:\9query9^^^^	Strongest overall reconstruction and holistic physics judgment
GPT-^^^^9 OR ti:\9^^^^ (Three.js)	PSNR ^^^^9max_results9query9^^^^.^^^^9 OR ti:\9ti:\9^^^^, SSIM ^^^^9query9^^^^.9^^^^9query9 OR ti:\9query9^^^^, success ^^^^9query9^^^^.99^^^^9query9 RAFT-EPE ^^^^9query9query9^^^^.^^^^9 OR ti:\9 OR ti:\9^^^^	High execution success and competitive motion alignment
Veo-^^^^9query9^^^^.^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^	PSNR ^^^^9max_results9query9^^^^.^^^^9query9ti:\9 SSIM ^^^^9query9^^^^.9^^^^9query9 OR ti:\9ti:\9^^^^, DINO ^^^^9query9^^^^.^^^^9 OR ti:\9 OR ti:\9query99^^^^, holistic ^^^^9max_results9^^^^.^^^^9 OR ti:\9max_results9^^^^	Strong semantic similarity without executable diagnosis
SVD img^^^^9max_results9^^^^vid	RAFT-EPE ^^^^9ti:\9 OR ti:\9^^^^.^^^^9ti:\9 OR ti:\9^^^^, holistic ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^.^^^^9ti:\9query9^^^^	Weaker motion reconstruction and physics plausibility

The qualitative error analysis is as important as the leaderboard. Models often reproduce object identities and layouts while assigning incorrect friction or restitution, producing premature stopping or excessive bouncing. Syntax and runtime failures still occur, though the single-retry repair mechanism improves success from ^^^^9query9^^^^.9^^^^9 OR ti:\99^^^^ to ^^^^9query9^^^^.99^^^^9query9 for Three.js and from ^^^^9query9^^^^.^^^^9 OR ti:\9 OR ti:\9query9^^^^ to ^^^^9query9^^^^.9^^^^9 OR ti:\99^^^^ for P^^^^9 OR ti:\9^^^^.js. The framework also identifies degenerate scenes—such as static or empty outputs—that can misleadingly improve motion metrics. For this reason, VisPhyWorld argues that RAFT-EPE must be read jointly with the holistic judge rather than in isolation (&&&^^^^9query9^^^^&&&).

^^^^9 OR ti:\9^^^^. Position within the broader physics-reasoning landscape

VisPhyWorld occupies a specific niche within recent work on physical reasoning: it evaluates whether a model can reconstruct an executable world hypothesis from observation. Nearby benchmarks and systems probe adjacent but distinct capabilities.

Work	Primary target	Distinguishing emphasis
VisPhyWorld (&&&^^^^9query9^^^^&&&)	Code-driven video reconstruction	Inspectable, editable, falsifiable simulator code
VideoPhy (&&&^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9^^^^&&&)	Physical commonsense in text-to-video	Binary Semantic Adherence and Physical Commonsense on ^^^^9 OR ti:\9 OR ti:\9 OR ti:\9^^^^ captions
ComPhy (&&&^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9ti:\9^^^^&&&)	Compositional physical reasoning from videos	Hidden mass and charge inference from reference videos
Phys^^^^9ti:\9^^^^D (&&&^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9 OR ti:\9^^^^&&&)	Physics-consistent ^^^^9ti:\9^^^^D world representations	Geometry-motion consistency and ^^^^9ti:\9^^^^D trajectory rewards
PhyEditBench (&&&^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9 OR ti:\9^^^^&&&)	Physics-aware image editing	Multi-stage real-world editing with ^^^^9max_results9query9 OR ti:\9^^^^ real and ^^^^9query9 OR ti:\9^^^^ Anti-Physics instances

VideoPhy evaluates whether generated videos follow physical commonsense for real-world activities, using two independent binary axes: Semantic Adherence and Physical Commonsense. Its human evaluation shows that current text-to-video models are far from accurate world simulation; the best joint performance reported in that benchmark is Pika at ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D99^^^^.^^^^9 OR ti:\9^^^^% for PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9 OR ti:\9^^^^ and PRESERVED_PLACEHOLDER_^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9 OR ti:\9^^^^ (&&&^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9query9^^^^&&&). Relative to VideoPhy, VisPhyWorld does not merely judge whether a video “looks physically right”; it demands an executable account of why.

ComPhy targets a different reasoning bottleneck: the inference of hidden intrinsic properties such as mass and electric charge from few-shot reference videos. The proposed Compositional Physics Learner achieves ^^^^9 OR ti:\9query9^^^^.^^^^9 OR ti:\9^^^^% factual accuracy, ^^^^9 OR ti:\9 OR ti:\9^^^^.^^^^9query9^^^^% predictive per-option accuracy / ^^^^9 OR ti:\9 OR ti:\9^^^^.^^^^9ti:\9^^^^% per-question, and ^^^^9 OR ti:\9 OR ti:\9^^^^.^^^^9query9^^^^% counterfactual per-option / ^^^^9max_results99^^^^.^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^% per-question, outperforming end-to-end video-QA baselines that struggle to use reference videos effectively (&&&^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9ti:\9^^^^&&&). This suggests a complementarity: ComPhy isolates latent property discovery, whereas VisPhyWorld externalizes physical hypotheses as runnable programs.

Phys^^^^9ti:\9^^^^D addresses physics consistency through explicit ^^^^9ti:\9^^^^D scene representations learned from video diffusion models. It uses a three-stage training paradigm—pseudo-supervised pretraining, physics-grounded supervised fine-tuning, and simulation-grounded reinforcement learning—and evaluates ^^^^9ti:\9^^^^D world consistency through geometric coherence, motion stability, and long-horizon plausibility. On its ^^^^9ti:\9^^^^D world evaluation, it reports ^^^^9ti:\9^^^^D Chamfer ^^^^9query9^^^^.^^^^9ti:\9 OR ti:\9max_results9 OR ti:\9^^^^, Worldline L^^^^9max_results9^^^^ Error ^^^^9query9^^^^.^^^^9ti:\99max_results9 OR ti:\9^^^^, Fail Rate 9.^^^^9 OR ti:\9ti:\9^^^^%, and Trajectory Length ^^^^9 OR ti:\9 OR ti:\9^^^^.^^^^9max_results9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9^^^^ (&&&^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9 OR ti:\9^^^^&&&). Compared with VisPhyWorld, Phys^^^^9ti:\9^^^^D emphasizes learned geometry-motion coupling rather than executable code synthesis.

PhyEditBench shifts the problem from video continuation or reasoning to instruction-based image editing under physical constraints. It defines ^^^^9ti:\9^^^^ primary classes and ^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9max_results9^^^^ subclasses, with ^^^^9max_results9query9 OR ti:\9^^^^ real-world instances and ^^^^9query9 OR ti:\9^^^^ synthetic Anti-Physics instances, and shows that current editing methods have substantial limitations in physics-based reasoning. Its training-free baseline PhyWorld leverages image-to-video generation as a reasoning mechanism for editing (&&&^^^^9VisPhyWorld physical reasoning code-driven video reconstruction (Liang et al., 9 Feb 2026) related VideoPhy ComPhy Phys4D9 OR ti:\9^^^^&&&). This line of work is adjacent to VisPhyWorld in its emphasis on causally coherent transitions, but its unit of evaluation is the edited image or transition sequence rather than the explicit simulator hypothesis.

Taken together, these works suggest that “physical reasoning” is not a single capability. It can mean hidden-property inference, commonsense judgment, world-consistent generation, physics-aware editing, or executable reconstruction. VisPhyWorld’s contribution is to make one of these dimensions—code-level causal commitment—directly testable.

^^^^9 OR ti:\9^^^^. Limitations, significance, and future directions

The current scope of VisPhyWorld is explicitly bounded. The benchmark is built from synthetic scenes, is predominately rigid-body, uses relatively short clips, and has moderate motion complexity with white backgrounds. The paper notes that ^^^^9query9^^^^D is harder, that current multimodal models struggle to autonomously generate stable code for more complex engines such as Unreal or Blender without human intervention, and that fluids, soft bodies, heavy occlusions, and long-horizon tasks are out of scope (&&&^^^^9query9^^^^&&&).

The framework also exposes several methodological trade-offs. Execution-based evaluation improves falsifiability, but it inherits the expressivity limits of the chosen simulator and renderer. Strong scene semantics do not imply correct parameterization. Non-physics renderers may superficially reconstruct layouts while failing mechanically. Holistic physics evaluation remains necessary because low-level motion alignment can be gamed by degenerate outputs. A plausible implication is that future systems will need stronger priors for ^^^^9query9^^^^D initialization, more capable repair loops, and explicit parameter-estimation metrics in addition to current reconstruction scores.

Within the broader literature, VisPhyWorld marks a shift from answer-based to hypothesis-based evaluation. In contrast to benchmarks that ask whether a model can label, rank, or narrate physical phenomena, it asks whether a model can instantiate them in code and survive execution. That makes the framework particularly relevant for applications where unverifiable physical claims are insufficient, including simulation-backed reasoning, controllable scene editing, and downstream embodied systems. Its core claim is therefore methodological as much as empirical: physical understanding becomes substantially more legible when it is forced into an executable, editable, and falsifiable world model (&&&^^^^9query9^^^^&&&).