PlaySuite is a large-scale benchmark that evaluates interactive visual intelligence by using over 5,000 open-source video games from platforms like PyWeek and itch.io. The benchmark covers diverse game engines (Pygame, HTML5, Godot, Unity) and introduces a unified closed-loop interaction framework and a Video-LLM-as-a-judge protocol to standardize progress measurement. Evaluation of fourteen recent models shows a perception-action gap, with strong reasoning but poor sustained progress, spatial grounding, action execution, and self-correction.
arXiv:2609.25001v1 Announce Type: new Abstract: Modern video games provide a measurable testbed for AI models, combining abilities of visual understanding, instruction decomposition, goal planning, a...
The paper introduces VISTA, a visual harness that equips a general-purpose multimodal model with long‑horizon vision and a lossless visual memory. VISTA enables the model to directly perceive and actively retrieve past observations, allowing it to reorganize visual input during reasoning. On the ARC‑AGI‑3 benchmark, VISTA boosts Claude Opus 5.0’s Relative Human Action Efficiency from 40.68 to a perfect 100.00, completing all 25 public games with 57.4% fewer actions than first‑time human participants, and it also outperforms baselines on three additional visual game and puzzle benchmarks.
WorldMind is a decoupled framework for state-aware NPC behavior in game world models, separating interactive world modeling into four layers: Understanding, Decision, Control, and Generation. It constructs a compact state from generated frames, reasons over it to plan NPC actions, translates actions into temporally aligned conditions, and synthesizes visual outcomes. Experiments on the newly introduced BOSS-140K dataset show that WorldMind achieves more tactically appropriate and coherent NPC behavior than baseline models in about 70% of pairwise comparisons.
The paper introduces VWG-Bench, a benchmark covering nine reasoning dimensions and 38 tasks to evaluate video generative models on symbolic reasoning, physical laws, and goal pursuit. It also presents Vid-PRE, a prompt-rewriting framework that offloads reasoning to a VLM, improving logical performance without changing the generator architecture. Experiments show that current models excel at visual quality but struggle with logic-heavy tasks, while Vid-PRE significantly boosts reasoning across multiple generators.
VBVR-Pro is a closed‑loop testbed that enables native visual reasoning through generation, offering 300 procedurally generated tasks that scale training and allow strong transfer to external benchmarks. It supplies verifiable reward scorers based on deterministic, task‑specific rules, outperforming VLM‑as‑a‑judge approaches and providing reliable signals for reinforcement learning. The suite also facilitates controlled modality studies, revealing that video generation excels at persistent spatiotemporal tracking while interleaved generation offers a compute‑efficient alternative, and highlights the importance of vision‑native trajectories for reasoning.
arXiv:2607. 01531v2 Announce Type: replace Abstract: Learning how an environment behaves from interaction is central to building agents that adapt to unfamiliar tasks.
SceneTeract is a verification interface that separates semantic action understanding from physical feasibility in indoor 3D scenes. It decomposes activities into atomic actions and performs explicit geometric checks to determine executability, providing diagnostic traces for failures. The system reveals widespread functional and accessibility issues in synthetic scenes, shows that existing VLMs over‑predict action feasibility, and improves VLM performance through post‑training with verifier feedback, with benefits that generalize to real‑world scenes.
AgentVidBench is a new multi‑hop video question‑answering benchmark designed to evaluate spatial, temporal, and causal reasoning in multimodal large language models (MLLMs). Unlike existing tests that focus on simple scene queries or global summaries, AgentVidBench includes step‑by‑step solution traces to assess whether agents gather the necessary evidence to justify their answers. Experiments with 12 MLLMs show limited single‑turn performance, but integrating these models into agentic workflows improves both accuracy and trajectory scores, establishing AgentVidBench as a comprehensive testbed for future research on agentic video understanding.
arXiv:2608.19583v2 Announce Type: replace-cross Abstract: Recent studies suggest that video generation models can exhibit certain forms of zero-shot visual reasoning through generated frames. Yet rel...
Recent studies suggest that video generation models can exhibit certain forms of zero-shot visual reasoning through generated frames. Yet reliable evaluation remains challenging: benchmarks should adopt inputs aligned with the visual priors of current video models, require valid evolving processes rather than only plausible final states, and calibrate task difficulty to remain challenging yet partly feasible.
PAVXploreRL introduces a reinforcement learning framework that builds on a pretrained latent world model to explicitly optimize Physical Plausibility, Action Adherence, and Visual Fidelity (PAV) objectives. By combining in‑distribution expert trajectories with noise‑driven out‑of‑distribution action exploration, the method avoids reliance on paired video supervision and improves generalization. Experiments demonstrate a 5.6% average performance gain over pretrained baselines and more reliable policy evaluation with reduced overestimation bias.
Building capable embodied agents requires not only multimodal perception and understanding, but also agentic capabilities for reasoning about actions, adapting to evolving situations, and interacting with the physical world. In this report, we introduce Hy-Embodied-VLM-1.