ORBIT: Scalable and Verifiable Data Generation for Search Agents on a Tight Budget
This addresses the problem of expensive data annotation for search agents, though it is incremental as it builds on existing synthetic data generation methods.
The authors tackled the challenge of constructing training datasets for search agents on complex queries by introducing ORBIT, a frugally generated dataset of 20K reasoning-intensive queries with verifiable answers, which when used to train a Qwen3-4B model achieved strong performance among sub-4B LLMs in search agent tasks.
Search agents, which integrate language models (LMs) with web search, are becoming crucial for answering complex user queries. Constructing training datasets for deep research tasks, involving multi-step retrieval and reasoning, remains challenging due to expensive human annotation, or cumbersome prerequisites. In this work, we introduce ORBIT, a training dataset with 20K reasoning-intensive queries with short verifiable answers, generated using a frugal framework without relying on paid API services. The modular framework relies on four stages: seed creation, question--answer pair generation, and two stages of verification: self and external. ORBIT spans 15 domains and each training pair requires 4--5 reasoning steps, with external search verification required from the complete web. We train Qwen3-4B as the base model on ORBIT using GRPO and evaluate it on Wikipedia question answering tasks. Extensive experiment results demonstrate that ORBIT-4B achieves strong performance among sub-4B LLMs as search agents, proving the utility of synthetic datasets. Our framework, code and datasets are open-sourced and available publicly.