{"slug": "pora-learning-when-more-reasoning-pays-off-for-llm-agents", "title": "PORA: Learning When More Reasoning Pays Off for LLM Agents", "summary": "A new arXiv paper (2610.27119v1) introduces Paired Outcome-guided Reasoning Allocation (PORA), a method that trains an LLM agent to decide before each action whether to spend a short or long reasoning budget. PORA trains by running each selected state twice, once with a short allowance and once with a long one, then compares final quality and total processing to estimate when longer reasoning pays off; at inference a controller weighs that estimate against learned prices on remaining-budget pressure without running both alternatives. On BFCL memory the method exceeds the strongest existing approach by 17.4–32.9 percentage points, and within the same agent it matches or beats always using long reasoning on code and mathematics with 50.6% fewer actor tokens and 31.2% fewer model requests, respectively.", "body_md": "arXiv:2610.27119v1 Announce Type: new \nAbstract: LLM agents perform tasks over multiple steps and revise their actions based on feedback. As the total reasoning budget is often limited and must support future steps, the agent must choose between short and long reasoning before each action. Longer reasoning may improve the final task outcome by preventing later errors, but it also uses budget capacity that may be needed for future decisions. This creates two challenges: estimating the downstream value of long reasoning and deciding when that value is worth using the remaining budget. We propose Paired Outcome-guided Reasoning Allocation (PORA), which learns this decision by comparing short and long reasoning from the same interaction state. During training, \\method{} completes the task twice from each selected state, once with a short reasoning allowance and once with a long allowance, and compares their final quality and total processing to estimate when longer reasoning is valuable. During inference, a controller weighs this estimate against learned prices on remaining-budget pressure and chooses either short or long reasoning without running both alternatives. We evaluate PORA across application, memory, code, retail, and mathematics tasks, achieving strong results across all settings. On BFCL memory, it exceeds the strongest existing method by 17.4--32.9 percentage points. Within the same agent, it matches or exceeds always using long reasoning on code and mathematics with 50.6% fewer actor tokens and 31.2% fewer model requests, respectively.", "url": "https://wpnews.pro/news/pora-learning-when-more-reasoning-pays-off-for-llm-agents", "canonical_source": "https://arxiv.org/abs/2610.27119", "published_at": "2026-10-21 04:00:00+00:00", "updated_at": "2026-10-21 04:20:33.575104+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-agents", "ai-research"], "entities": ["Paired Outcome-guided Reasoning Allocation", "PORA", "BFCL", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/pora-learning-when-more-reasoning-pays-off-for-llm-agents", "markdown": "https://wpnews.pro/news/pora-learning-when-more-reasoning-pays-off-for-llm-agents.md", "text": "https://wpnews.pro/news/pora-learning-when-more-reasoning-pays-off-for-llm-agents.txt", "jsonld": "https://wpnews.pro/news/pora-learning-when-more-reasoning-pays-off-for-llm-agents.jsonld"}}