{"slug": "in-model-context-compression-with-agentic-soft-tokens-for-long-horizon-llm", "title": "In-Model Context Compression with Agentic Soft Tokens for Long-Horizon LLM Agents", "summary": "A paper posted to arXiv (2610.41291v1) introduces Agentic Soft-Token Compression (ASTC), a training-free method that compresses long-horizon LLM agent context using soft tokens inside the model rather than external compressors. ASTC uses the model's native recurrent dynamics via hybrid Gated Delta Network (GDN)-attention to identify already-absorbed information, carrying it forward as agentic soft tokens while retaining exact values as hard tokens, which reduces later-layer computation and yields reusable KV caches. In long-horizon experiments, ASTC matched or exceeded the uncompressed agent's accuracy with up to 50.68% fewer end-to-end Attention Pairs, while trained compressors required 15.4 to 103.7 times more Attention Pairs per successful task.", "body_md": "arXiv:2610.41291v1 Announce Type: new \nAbstract: Long-horizon LLM agents play an important role in solving complex and sustained tasks, with ever-growing interaction contexts. However, these accumulated contexts will raise substantial computational cost and obscure key decision-relevant evidence. Existing methods typically train external compressors or context-management agents to rewrite context as shorter hard-token sequences, but such external selection may not align with the LLM's evolving representation and is necessarily lossy. In this paper, we introduce Agentic Soft-Token Compression (ASTC), a training-free compression method using soft tokens within LLM agent for long-horizon tasks. Specifically, ASTC uses the model's native recurrent dynamics to determine the already absorbed information, leveraging the commonly-used hybrid Gated Delta Network (GDN)-attention. It then carries this information forward as agentic soft tokens while retaining exact values as hard tokens. The resulting compact sequence reduces later-layer computation and yields reusable KV caches, extending the savings across the agent's trajectory. Extensive experiments on long-horizon tasks show that ASTC matches or exceeds the uncompressed agent's accuracy with up to 50.68% fewer end-to-end Attention Pairs, while trained compressors require 15.4 to 103.7 times more Attention Pairs per successful task. These findings highlight agentic soft tokens as a model-native way to carry absorbed context forward as the task representation evolves.", "url": "https://wpnews.pro/news/in-model-context-compression-with-agentic-soft-tokens-for-long-horizon-llm", "canonical_source": "https://arxiv.org/abs/2610.41291", "published_at": "2026-11-02 05:00:00+00:00", "updated_at": "2026-11-02 05:20:18.023247+00:00", "lang": "en", "topics": ["large-language-models", "ai-agents", "ai-research", "artificial-intelligence", "machine-learning"], "entities": ["arXiv", "Agentic Soft-Token Compression", "Gated Delta Network"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/in-model-context-compression-with-agentic-soft-tokens-for-long-horizon-llm", "markdown": "https://wpnews.pro/news/in-model-context-compression-with-agentic-soft-tokens-for-long-horizon-llm.md", "text": "https://wpnews.pro/news/in-model-context-compression-with-agentic-soft-tokens-for-long-horizon-llm.txt", "jsonld": "https://wpnews.pro/news/in-model-context-compression-with-agentic-soft-tokens-for-long-horizon-llm.jsonld"}}