SFR-DeepResearch: Towards Effective Reinforcement Learning for Autonomously Reasoning Single Agents
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908526755971072 |
|---|---|
| author | Nguyen, Xuan-Phi Pandit, Shrey Reddy, Revanth Gangi Xu, Austin Savarese, Silvio Xiong, Caiming Joty, Shafiq |
| author_facet | Nguyen, Xuan-Phi Pandit, Shrey Reddy, Revanth Gangi Xu, Austin Savarese, Silvio Xiong, Caiming Joty, Shafiq |
| contents | Equipping large language models (LLMs) with complex, interleaved reasoning and tool-use capabilities has become a key focus in agentic AI research, especially with recent advances in reasoning-oriented (``thinking'') models. Such capabilities are key to unlocking a number of important applications. One such application is Deep Research (DR), which requires extensive search and reasoning over many sources. Our work in this paper focuses on the development of native Autonomous Single-Agent models for DR featuring minimal web crawling and Python tool integration. Unlike multi-agent systems, where agents take up pre-defined roles and are told what to do at each step in a static workflow, an autonomous single-agent determines its next action dynamically based on context, without manual directive. While prior work has proposed training recipes for base or instruction-tuned LLMs, we focus on continual reinforcement learning (RL) of reasoning-optimized models to further enhance agentic skills while preserving reasoning ability. Towards this end, we propose a simple RL recipe with entirely synthetic data, which we apply to various open-source LLMs. Our best variant SFR-DR-20B achieves up to 28.7% on Humanity's Last Exam benchmark. In addition, we conduct key analysis experiments to provide more insights into our methodologies. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_06283 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | SFR-DeepResearch: Towards Effective Reinforcement Learning for Autonomously Reasoning Single Agents Nguyen, Xuan-Phi Pandit, Shrey Reddy, Revanth Gangi Xu, Austin Savarese, Silvio Xiong, Caiming Joty, Shafiq Artificial Intelligence Computation and Language Equipping large language models (LLMs) with complex, interleaved reasoning and tool-use capabilities has become a key focus in agentic AI research, especially with recent advances in reasoning-oriented (``thinking'') models. Such capabilities are key to unlocking a number of important applications. One such application is Deep Research (DR), which requires extensive search and reasoning over many sources. Our work in this paper focuses on the development of native Autonomous Single-Agent models for DR featuring minimal web crawling and Python tool integration. Unlike multi-agent systems, where agents take up pre-defined roles and are told what to do at each step in a static workflow, an autonomous single-agent determines its next action dynamically based on context, without manual directive. While prior work has proposed training recipes for base or instruction-tuned LLMs, we focus on continual reinforcement learning (RL) of reasoning-optimized models to further enhance agentic skills while preserving reasoning ability. Towards this end, we propose a simple RL recipe with entirely synthetic data, which we apply to various open-source LLMs. Our best variant SFR-DR-20B achieves up to 28.7% on Humanity's Last Exam benchmark. In addition, we conduct key analysis experiments to provide more insights into our methodologies. |
| title | SFR-DeepResearch: Towards Effective Reinforcement Learning for Autonomously Reasoning Single Agents |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2509.06283 |