The paper "Context Compaction for World Models" introduces a benchmark to evaluate how well world models retrieve context when revisiting regions, and presents an agent harness that compacts context to extend retrieval beyond the model’s native window. It demonstrates that this compaction allows models to recall rooms up to four times farther than their default context size without retraining, while models using sliding‑window attention lose most of their retrieval capability during dreaming. The study compares four world‑model designs, showing that extending context reach via compaction benefits dreaming agents more than switching to sliding‑window attention.
arXiv:2607. 23809v1 Announce Type: new Abstract: Agentic tasks are inherently long-horizon and multi-turn, constantly accumulating context through interactions with the environment.
The paper evaluates two strategies for training language‑model‑based world models—fine‑tuning and Retrieval Augmented Generation (RAG)—across five diverse environments. Fine‑tuning generally yields higher rewards, while RAG is more data‑efficient. The authors propose a counterfactual error estimation for RAG, improve retrieval with a hierarchical query reformulation, and combine both approaches into a hybrid system that outperforms each method alone.
The paper introduces CTXAgent, a self‑controlled context management system for large language model agents that compacts conversation history when a token threshold is reached. It evaluates seven frontier models, comparing their ability to manage context with and without dedicated tools, and benchmarks performance across coding, scientific, and software engineering tasks. Results show that well‑managed models keep live context below 100k tokens, maintain high benchmark scores, and achieve significant cost reductions, with Claude Opus 5.5 outperforming its explicit configuration by 34% lower cost.
arXiv:2606. 00408v1 Announce Type: cross Abstract: Long-horizon search agents accumulate large amounts of retrieved content across many tool calls, making context-budget efficiency increasingly important.
The paper introduces Learning-to-Retrieve (L2R), a method that embeds memory retrieval directly into video world models. By using the model’s internal state as a memory and employing a camera‑conditioned retrieval gate and a retrieval trigger supervised by a 3D re‑visibility signal, the approach learns when and what historical information to access. This eliminates the need for external memory systems and improves long‑term scene consistency across various base models and camera‑revisit benchmarks.
arXiv:2609.23121v1 Announce Type: new Abstract: Multimodal Agentic Retrieval (MAR) requires agents to solve complex information-seeking tasks by iteratively invoking external tools. Typical framework...
arXiv:2608.21690v1 Announce Type: new Abstract: LLM agents increasingly take on long-running tasks whose history grows far beyond a single model context window. Existing approaches compress earlier i...
WorldAgen is a unified framework that jointly learns world modeling and action prediction using a shared Transformer backbone with two specialized heads. It introduces a Mixed Unidirectional Attention Mask to separate the world model and agent model, and enables Test-Time Training (TTT) by sampling exploratory actions and updating the world model with real state transitions. Experiments on CALVIN and LIBERO show that WorldAgen matches or surpasses state‑of‑the‑art methods, especially when TTT is applied to a few samples.
arXiv:2609.40118v1 Announce Type: new Abstract: As LLM capabilities advance, agents are tackling increasingly complex tasks over longer horizons. Their growing interaction histories make memory compa...
arXiv:2610.28554v1 Announce Type: new Abstract: Despite the success of video world models, extending them to multiple agents is difficult because it requires generating several agent-specific video s...
arXiv:2608. 07408v1 Announce Type: cross Abstract: We study visual persistence in interactive video world models.
MemoWM proposes a memory allocation framework that leverages world models to compress what long‑term agents retain. By using shared predictions, it reconstructs omitted content and balances reconstruction error against storage cost, keeping only information that adds downstream value beyond the predictive prior. On five long‑term agent‑memory benchmarks, MemoWM improves average answer accuracy by 42.42% and cuts experience‑specific storage by 53.9% compared to the most storage‑efficient baseline.