Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces AutoWorldModel-Bench, a novel benchmark designed for evaluating autonomous coding agents in the realm of world modeling, where traditional methods fall short due to the complex interactions among architectures, training objectives, and state representations. By providing a structured-state representation across eight game environments, the benchmark allows agents to iteratively improve a world-model starter within a fixed compute budget, fostering open-ended research rather than mere engineering tasks. Results show that agents like Codex-5.4 and Claude Opus 4.6 successfully enhance their models in 63 out of 64 sessions, with significant modifications often involving new objectives or architectural changes rather than simple hyperparameter adjustments.
Autonomous coding agents can outperform traditional world-modeling approaches by making substantial, innovative modifications rather than just fine-tuning parameters.
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 world-model starter 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 starter on 63; in 91% of sessions the winning edit is a non-trivial research-style modification--a new objective, representation, rollout procedure, or architectural change--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.