QUEST: Training Frontier Deep Research Agents with Fully Synthetic Tasks

Deep research agents extend the role of search engines from retrieving keyword-matched pages to synthesizing knowledge, fundamentally changing how humans interact with information. However, frontier systems remain proprietary, while existing open agents often generalize poorly across different task types, leaving unclear how to train a broadly capable deep research agent. We release Quest, a family of open models (ranging from 2B to 35B) that serve as general-purpose deep research agents designed to handle a wide range of long-horizon search tasks, with strong capabilities in fact seeking, citation grounding, and report synthesis. To build Quest, we propose an effective training recipe combining mid-training, supervised fine-tuning, and reinforcement learning. Central to this recipe is a curated data synthesis pipeline based on unified rubric trees, which applies to different task types and enables synthesizing training data with verifiable rewards without human annotation. In addition, Quest incorporates a built-in context management mechanism that enables effective longhorizon reasoning and knowledge synthesis.

Introduction. Web search is moving from manual information gathering toward autonomous evidence-seeking and synthesis. This shift builds on a progression of search interfaces: traditional search engines return ranked webpages for users to inspect, while retrieval-augmented generation (RAG) systems (Lewis et al., 2020; Gao et al., 2023) retrieve relevant documents and condition LLM responses on them. Deep research agents (OpenAI, 2025c; Google DeepMind, 2025; Tongyi et al., 2025) push this paradigm further: given a complex information-seeking task, they can decompose it into intermediate goals, execute web queries, examine external sources, and synthesize citation-backed responses. By shifting the burden of web-scale information seeking from humans to autonomous agents, these systems promise to make complex research workflows more efficient and scalable. Yet building such agents is far from straightforward: they must learn not only to answer questions, but to search, verify, remember, and synthesize information over long horizons.

Discussion / Conclusion. Quest is an open family of general-purpose deep research agents capable of solving long-horizon tasks that require fact seeking, report synthesis, and citation grounding. The largest model, Quest-35B, achieves the best overall performance among recent open-weight agents and approaches or even surpasses closed-source agents across eight benchmarks. Beyond the models themselves, the central contribution is an open, reproducible recipe: rubric-tree-based data synthesis, structured context management, efficient training infrastructure, and a staged training pipeline combining mid-training, supervised finetuning, and RL. We hope this work provides a strong open foundation for future deep research agents.