OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis

Papers with Code Trending Papers
deep-research agent-training trajectory-synthesis open-source fine-tuning search-browse llm

Summary

OpenResearcher presents a reproducible pipeline for training deep research agents using offline search environments and synthesized trajectories, achieving significant accuracy improvements on benchmark tasks like BrowseComp-Plus.

Training deep research agents requires long-horizon trajectories that interleave search, evidence aggregation, and multi-step reasoning. However, existing data collection pipelines typically rely on proprietary web APIs, making large-scale trajectory synthesis costly, unstable, and difficult to reproduce. We present OpenResearcher, a reproducible pipeline that decouples one-time corpus bootstrapping from multi-turn trajectory synthesis and executes the search-and-browse loop entirely offline using three explicit browser primitives: search, open, and find, over a 15M-document corpus. Using GPT-OSS-120B as the teacher model, we synthesize over 97K trajectories, including a substantial long-horizon tail with 100+ tool calls. Supervised fine-tuning a 30B-A3B backbone on these trajectories achieves 54.8\% accuracy on BrowseComp-Plus, a +34.0 point improvement over the base model, while remaining competitive on BrowseComp, GAIA, and xbench-DeepSearch. Because the environment is offline and fully instrumented, it also enables controlled analysis, where our study reveals practical insights into deep research pipeline design, including data filtering strategies, agent configuration choices, and how retrieval success relates to final answer accuracy. We release the pipeline, synthesized trajectories, model checkpoints, and the offline search environment at https://github.com/TIGER-AI-Lab/OpenResearcher.
Original Article
View Cached Full Text

Cached at: 07/16/26, 10:24 AM

Paper page - OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis

Source: https://huggingface.co/papers/2603.20278

Abstract

OpenResearcher presents a reproducible pipeline for training deep research agents using offline search environments and synthesized trajectories, achieving improved accuracy on benchmark tasks.

Trainingdeep research agentsrequireslong-horizon trajectoriesthat interleave search, evidence aggregation, and multi-step reasoning. However, existing data collection pipelines typically rely on proprietary web APIs, making large-scale trajectory synthesis costly, unstable, and difficult to reproduce. We present OpenResearcher, a reproducible pipeline that decouples one-time corpus bootstrapping from multi-turn trajectory synthesis and executes thesearch-and-browse loopentirely offline using three explicitbrowser primitives: search, open, and find, over a 15M-document corpus. UsingGPT-OSS-120Bas the teacher model, we synthesize over 97K trajectories, including a substantial long-horizon tail with 100+ tool calls.Supervised fine-tuninga30B-A3B backboneon these trajectories achieves 54.8\% accuracy onBrowseComp-Plus, a +34.0 point improvement over the base model, while remaining competitive onBrowseComp,GAIA, andxbench-DeepSearch. Because the environment is offline and fully instrumented, it also enables controlled analysis, where our study reveals practical insights into deep research pipeline design, including data filtering strategies, agent configuration choices, and how retrieval success relates to final answer accuracy. We release the pipeline, synthesized trajectories, model checkpoints, and the offline search environment at https://github.com/TIGER-AI-Lab/OpenResearcher.

View arXiv pageView PDFProject pageGitHub970Add to collection

Get this paper in your agent:

hf papers read 2603\.20278

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper2

#### OpenResearcher/OpenResearcher-30B-A3B Text Generation• 32B• UpdatedMar 25 • 1.43k • 69 #### OpenResearcher/OpenResearcher-30B-A3B-GGUF 32B• UpdatedMar 24 • 190 • 13

Datasets citing this paper7

#### i-DeepSearch/observation-masking-eval-logs Preview• Updated1 day ago • 9.99k • 2 #### OpenResearcher/OpenResearcher-Dataset Viewer• UpdatedMar 25 • 97.6k • 6.65k • 130 #### OpenResearcher/web-bench Viewer• UpdatedMay 19 • 5.5k • 4.48k • 4 #### OpenResearcher/OpenResearcher-Corpus Viewer• UpdatedMar 25 • 14.9M • 1.14k • 10 Browse 7 datasets citing this paper### Spaces citing this paper7

Collections including this paper13

Browse 13 collections that include this paper

Similar Articles

RODS: Reward-Driven Online Data Synthesis for Multi-Turn Tool-Use Agents

arXiv cs.AI

This paper introduces RODS, a reward-driven online data synthesis method that addresses the depletion of informative samples in static datasets for multi-turn tool-use agent training. It achieves comparable performance to larger offline pipelines with significantly fewer trajectories.