Benchmarking World-Model Learning

• URL: http://arxiv.org/abs/2510.19788v2
• Date: Thu, 23 Oct 2025 11:48:07 GMT
• Title: Benchmarking World-Model Learning
• Abstract summary: We present WorldTest, a protocol to evaluate model-learning agents that separates a scored phase in a different but related environment.<n>We compared 517 human participants and three models on AutumnBench.<n>We found that humans outperform the models in some environments but not others.
• Score: 48.64929865461133
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Model-learning agents should gather information to learn world models that support many downstream tasks and inferences, such as predicting unobserved states, estimating near- and far-term consequences of actions, planning action sequences, and detecting changes in dynamics. Current methods for learning and evaluating world models diverge from this goal: training and evaluation are anchored to next-frame prediction, and success is scored by reward maximization in the same environment. We propose WorldTest, a protocol to evaluate model-learning agents that separates reward-free interaction from a scored test phase in a different but related environment. WorldTest is open-ended$\unicode{x2014}$models should support many different tasks unknown ahead of time$\unicode{x2014}$and agnostic to model representation, allowing comparison across approaches. We instantiated WorldTest with AutumnBench, a suite of 43 interactive grid-world environments and 129 tasks across three families: masked-frame prediction, planning, and predicting changes to the causal dynamics. We compared 517 human participants and three frontier models on AutumnBench. We found that humans outperform the models, and scaling compute improves performance only in some environments but not others. WorldTest provides a novel template$\unicode{x2014}$reward-free exploration, derived tests, and behavior-based scoring$\unicode{x2014}$to evaluate what agents learn about environment dynamics, and AutumnBench exposes significant headroom in world-model learning.

Related papers

• WorldAgen: Unified State-Action Prediction with Test-Time World Model Training [19.89058127984308]
  We present WorldAgen, a unified framework that jointly learns world modeling and action prediction.<n>WorldAgen samples exploratory actions, collects ground-truth state transitions, and performs lightweight TTT updates to refine its world model.<n>Our method surpasses existing state-of-the-art models, highlighting the effectiveness of adapting world models at inference time.
  arXiv  Detail & Related papers  (2026-09-08T02:46:13Z)
• Discriminative World Models for Web Agents [52.31107486410659]
  Recent web agents use world models for test-time action selection by sampling candidate actions, predicting the resulting web states, and ranking them with a ranker model or a Process Reward Model (PRM)<n>We introduce predicted-state matching, a training objective where the predicted representation must distinguish the true resulting state from those reached by alternative actions.<n>Our approach improves PRM-style action ranking on WebPRMBench compared with action-only PRMs and PRMs augmented with supervised-next-state world models.
  arXiv  Detail & Related papers  (2026-09-02T17:59:40Z)
• iWorld-Bench: A Benchmark for Interactive World Models with a Unified Action Generation Framework [27.208236690012914]
  iWorld-Bench is a benchmark for training and testing world models on interaction-related abilities.<n>We construct a diverse dataset with 330k video clips and select 2.1k high-quality samples covering varied perspectives, weather, and scenes.
  arXiv  Detail & Related papers  (2026-05-05T16:30:03Z)
• Beyond State Consistency: Behavior Consistency in Text-Based World Models [58.216587360435305]
  We introduce a new behavior-aligned training paradigm aimed at improving the functional consistency between the world model and the real environment.<n> Experiments on WebShop and TextWorld show that BehR-based training improves long-term alignment in several settings.
  arXiv  Detail & Related papers  (2026-04-15T12:56:45Z)
• MMEarth-Bench: Global Model Adaptation via Multimodal Test-Time Training [15.675086189757769]
  MMEarth-Bench is a collection of five new multimodal environmental tasks with 12 modalities, globally distributed data, and both in- and out-of-distribution test splits.<n>We benchmark a diverse set of pretrained models and find that while (multimodal) pretraining tends to improve model robustness in limited data settings, geographic generalization abilities remain poor.<n>We propose a model-agnostic method for test-time training with multimodal reconstruction (TTT-MMR) that uses all the modalities available at test time as auxiliary tasks.
  arXiv  Detail & Related papers  (2026-02-06T00:48:19Z)
• Agent2World: Learning to Generate Symbolic World Models via Adaptive Multi-Agent Feedback [51.22403664895878]
  Agent2World is a tool-augmented multi-agent framework that achieves strong inference-time world-model generation.<n>It also serves as a data engine for supervised fine-tuning, by grounding generation in multi-agent feedback.
  arXiv  Detail & Related papers  (2025-12-26T18:54:14Z)
• Bootstrapping World Models from Dynamics Models in Multimodal Foundation Models [37.774994737939394]
  We use dynamics models to bootstrap world models using synthetic data and inference time verification.<n>Our best model achieves a performance competitive with state-of-the-art image editing models, improving on them by a margin of $15%$ on real-world subsets according to GPT4o-as-judge.
  arXiv  Detail & Related papers  (2025-06-06T11:50:18Z)
• WorldPrediction: A Benchmark for High-level World Modeling and Long-horizon Procedural Planning [52.36434784963598]
  We introduce WorldPrediction, a video-based benchmark for evaluating world modeling and procedural planning capabilities of different AI models.<n>We show that current frontier models barely achieve 57% accuracy on WorldPrediction-WM and 38% on WorldPrediction-PP whereas humans are able to solve both tasks perfectly.
  arXiv  Detail & Related papers  (2025-06-04T18:22:40Z)
• Exploration-Driven Generative Interactive Environments [53.05314852577144]
  We focus on using many virtual environments for inexpensive, automatically collected interaction data.<n>We propose a training framework merely using a random agent in virtual environments.<n>Our agent is fully independent of environment-specific rewards and thus adapts easily to new environments.
  arXiv  Detail & Related papers  (2025-04-03T12:01:41Z)
• Can foundation models actively gather information in interactive environments to test hypotheses? [43.42688356541211]
  Foundation models excel at single-turn reasoning but struggle with multi-turn exploration in dynamic environments.<n>We evaluated these models on their ability to learn from experience, adapt, and gather information.
  arXiv  Detail & Related papers  (2024-12-09T12:27:21Z)
• STORM: Efficient Stochastic Transformer based World Models for Reinforcement Learning [82.03481509373037]
  Recently, model-based reinforcement learning algorithms have demonstrated remarkable efficacy in visual input environments. We introduce Transformer-based wORld Model (STORM), an efficient world model architecture that combines strong modeling and generation capabilities. Storm achieves a mean human performance of $126.7%$ on the Atari $100$k benchmark, setting a new record among state-of-the-art methods.
  arXiv  Detail & Related papers  (2023-10-14T16:42:02Z)
• A Control-Centric Benchmark for Video Prediction [69.22614362800692]
  We propose a benchmark for action-conditioned video prediction in the form of a control benchmark. Our benchmark includes simulated environments with 11 task categories and 310 task instance definitions. We then leverage our benchmark to study the effects of scaling model size, quantity of training data, and model ensembling.
  arXiv  Detail & Related papers  (2023-04-26T17:59:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.