This work introduces AutoWorldModel-Bench, a closed-loop benchmark in which frontier coding agents autonomously improve a provided base world model under a fixed compute budget, and offers a setting in which frontier coding agents can be evaluated on open-ended research rather than engineering-to-spec problems.
Abstract
World modeling is an unsettled field: architectures, training objectives, and state representations interact in complex ways, and no single recipe dominates across environments. This makes it an ideal testbed for AI coding agents acting as autonomous researchers--a setting in which the improvement direction is not specified in advance, unlike the engineering-to-spec tasks that dominate current agent benchmarks. We introduce AutoWorldModel-Bench, a closed-loop benchmark in which frontier coding agents autonomously improve a provided base world model under a fixed compute budget. The benchmark spans eight game environments under a unified structured-state representation--ground-truth entity state extracted from each game and consumed through a shared tensor format--which isolates dynamics modeling from perception and enables minutes-per-run iteration. Across 64 sessions, Codex-5.4 and Claude Opus 4.6 improve their base on a held-out test split in all but one session, with about half (33 of 64) a substantial gain ($\Delta \geq +0.10$) and the remaining improvements smaller but positive; in 91% of sessions the winning edit is a substantive change to the model or training rather than a hyperparameter tweak. Our benchmark offers a setting in which frontier coding agents can be evaluated on open-ended research rather than engineering-to-spec problems.
VisualPatchWorld is introduced, which represents world dynamics as code and first selects a qualitative dynamical form with short active probes, then fits that form's free parameters from recorded state-action traces by minimizing multi-step prediction error.
Latent world-action models avoid rendering future pixels by predicting an action-relevant visual subgoal in feature space. LaWAM established this formulation, but its original presentation left the world model, multimodal backbone, and deployment checkpoint tightly coupled. We introduce AcrossWAM1.0, a modularization and scaling study of this latent world-action stack. Rather than presenting latent subgoals as a new algorithm, we make the module boundary explicit: a policy adapter produces latent-action and action-generation contexts; a retained latent world decoder grounds the predicted transition in the current scene;and a flow-matching expert generates continuous action chunks. We further separate training-only teachers from the inference graph and provide a verifiable deployment export. On 2,000 paired LIBERO episodes, replacing a Qwen3-VL-2B backbone with Qwen3.5-0.8B yields 97.45% success versus 98.00% for the 2B model (a-0.55percentage-point difference; exact McNemarp=0.266). This does not prove equivalence, but it meets a prespecified two-point retention criterion. The compact, inference-reachable checkpoint contains 1,472.6M unique parameters, 42.4% fewer than the original 2B policy, while all retained tensors are bitwise identical to the source checkpoint. Cross-family execution is additionally checked with a MiniCPM-V adapter smoke test; closed-loop cross-family transfer remains an open evaluation. AcrossWAM1.0 therefore contributes an auditable software and evaluation boundary for compact latent world-action policies, distinct from LaWAM's original latent-subgoal contribution.
This work presents Echoverse, which compiles specifications into stateful applications whose tasks are graded against the application's own database, and a co-evolution loop that reads every graded rollout twice: as repairs to the environment, its tasks and its verifier, and as training signal for the model.
This work conceptualize Agent-Centric Interactive World Proxies, shifting the fundamental paradigm from physical state transitions to agent-usable information transitions, such as execution outcomes, retrieved experiences or skills, and verification signals, broadening the scope of world modeling to provide versatile feedback for continually improving agents.
Yu Yang, Xuemeng Yang, Licheng Wen et al.· 0 citations
Large-language-model (LLM) agents perform well in embodied benchmarks but are costly, stochastic, and difficult to audit. We propose an LLM-free zero-shot decision agent for ALFWorld that combines structured commonsense priors with adaptive calibration. The agent uses object-location priors, task templates, synonym mappings, and a hierarchical state controller over admissible commands. Because every decision is traceable to explicit knowledge entries, success and failure signals update only the responsible entries rather than all parameters. On 134 ALFWorld valid_unseen tasks, static priors obtain 67.2% success; symmetric calibration over P(obj,loc), M(target,entity), and S(word) improves this to 73.9%, with one-round convergence and CPU-only execution. We also observe that 34/134 tasks contain description-environment inconsistencies; on the consistent subset, our system reaches 93.0%. The results show that interpretable structured priors can be a practical alternative for well-specified embodied decision making.
Shengjie Ma, Jin-Han Li, Guo-An Zhang et al.· 2026 3rd World Conference on...· 0 citations
Desc descriptive evidence is provided that long-horizon multi-tool post-training can change ways of working that transfer beyond its training domain, and both software-engineering benchmarks improve despite the training collection containing no software-engineering tasks.
Sushant Mehta, Logan Ritchie, Liudas Panavas et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
