{"version":1,"type":"story","url":"https://digestai.news/story/arxiv-study-finds-ptr-outperforms-rtp-for-multi-tool-queries","json":"https://digestai.news/story/arxiv-study-finds-ptr-outperforms-rtp-for-multi-tool-queries.json","markdown":"https://digestai.news/story/arxiv-study-finds-ptr-outperforms-rtp-for-multi-tool-queries.md","slug":"arxiv-study-finds-ptr-outperforms-rtp-for-multi-tool-queries","headline":"arXiv study finds PtR outperforms RtP for multi-tool queries","summary":"A new paper on arXiv compares two approaches for tool-augmented AI agents: Retrieve-then-Plan (RtP) and Plan-then-Retrieve (PtR). The study uses ToolBench, a dataset with 46,984 unique API endpoints across 50 categories, to test both methods on single-tool and multi-tool tasks.\n\nResearchers found that RtP works better for simple, single-tool queries, while PtR excels on complex, multi-tool scenarios. PtR’s advantage persists across different retrieval budgets (from 1 to 20 results) and registry sizes (ranging from 1,000 to 47,000 APIs). The paper also notes that PtR’s extra discovery step roughly doubles inference costs, but this trade-off benefits multi-tool tasks. The authors highlight a failure mode where PtR’s generated tool stubs may hallucinate non-existent API names, which can hurt performance on simpler queries but help break down complex intents.","keyPoints":["PtR outperforms RtP on multi-tool queries but underperforms on single-tool tasks across all tested conditions","Study tested both methods on ToolBench with 46,984 APIs and 50 categories, using models like Gemini 2.5 Flash","PtR’s extra retrieval step doubles inference cost but may be worth it for complex multi-tool scenarios"],"whyItMatters":"The findings suggest AI agents could adapt their tool-discovery strategy based on query complexity, improving efficiency and reducing unnecessary costs.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":[],"models":["Gemini 2.5 Flash","gpt-oss-120b"],"people":[]},"firstPublishedAt":"2026-10-15T04:00:00Z","updatedAt":"2026-10-15T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.AI","title":"Plan or Retrieve First? A Scale Study of Tool Discovery Architectures","url":"https://arxiv.org/abs/2610.19068","publishedAt":"2026-10-15T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"arXiv study finds PtR outperforms RtP for multi-tool queries\", 15 October 2026, https://digestai.news/story/arxiv-study-finds-ptr-outperforms-rtp-for-multi-tool-queries","publisher":"Digest AI","title":"arXiv study finds PtR outperforms RtP for multi-tool queries","datePublished":"2026-10-15T04:00:00Z","url":"https://digestai.news/story/arxiv-study-finds-ptr-outperforms-rtp-for-multi-tool-queries"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}