aiqbench analysis
AgentOCR addresses the critical bottleneck of growing textual histories in agentic systems by converting observation-action logs into compact rendered images. This approach leverages visual tokens to reduce token consumption by over 50% while maintaining high task success rates through a novel self-compression mechanism. The framework demonstrates significant scalability improvements for multi-turn reinforcement learning deployments without sacrificing performance.
AgentOCR: Reimagining Agent History via Optical Self-Compression
arXiv:2601.04786v3 Announce Type: replace-cross Abstract: Recent advances in large language models (LLMs) enable agentic systems trained with reinforcement learning (RL) over multi-turn interaction, but practical deployment is bottlenecked by rapidly growing textual histories that inflate token and memory costs. We introduce AgentOCR, a framework that exploits visual tokens' superior information density by representing the accumulated observation-action history as a compact rendered image. To make multi-turn rollouts scalable, AgentOCR proposes segment optical caching. By decomposing history into hashable segments and maintaining a visual cache, this mechanism eliminates redundant re-rendering. Beyond fixed rendering, AgentOCR introduces agentic self-compression, where the agent actively emits a compression rate and is trained with compression-aware reward to adaptively balance task success and token efficiency. We conduct extensive experiments on challenging agentic benchmarks, ALFWorld and search-based QA. Remarkably, AgentOCR preserves over 95\% of text-based agent performance while substantially reducing token consumption (>50\%), yielding consistent token and memory efficiency. Further analysis validates a 20$\times$ rendering speedup from optical caching and effective self-compression balancing. Our code is available at https://github.com/langfengQ/AgentOCR.
Abstract:Recent advances in large language models (LLMs) enable agentic systems trained with reinforcement learning (RL) over multi-turn interaction, but practical deployment is bottlenecked by rapidly growing textual histories that inflate token and memory costs. We introduce AgentOCR, a framework that exploits visual tokens' superior information density by representing the accumulated observation-action history as a compact rendered image. To make multi-turn rollouts scalable, AgentOCR proposes segment optical caching. By decomposing history into hashable segments and maintaining a visual cache, this mechanism eliminates redundant re-rendering. Beyond fixed rendering, AgentOCR introduces agentic self-compression, where the agent actively emits a compression rate and is trained with compression-aware reward to adaptively balance task success and token efficiency. We conduct extensive experiments on challenging agentic benchmarks, ALFWorld and search-based QA. Remarkably, AgentOCR preserves over 95\% of text-based agent performance while substantially reducing token consumption (>50\%), yielding consistent token and memory efficiency. Further analysis validates a 20$\times$ rendering speedup from optical caching and effective self-compression balancing. Our code is available at this https URL.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2601.04786 [cs.LG]
(or arXiv:2601.04786v3 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2601.04786 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 262 words
Related coverage
Reading is not Reasoning: Bridging the Agentic Policy Gap in Vision-Text Compression
arXiv · Oct 15, 2026
When History Is Multimodal: Rethinking Context Management for Long-Horizon Agents
arXiv · Sep 1, 2026
Game Design Creativity evals
LessWrong · Oct 12, 2026
[AINews] not much happened today
Latent Space · Sep 18, 2026