AI(X) Hub at The Ohio State University

Amazon – The Ohio State University – Deep Research Partnership: 2.5 Million Views in Under 24 Hours

On the evening of June 17, 2026, Ohio State University's NLP Group, in collaboration with Amazon, posted QUEST — an open-source deep research AI agent that matches or surpasses proprietary systems from OpenAI and Anthropic. Within hours, the announcement went viral on X (formerly Twitter), accumulating over 2.5 million views and counting.
June 23, 2026

Amazon – The Ohio State University – Deep Research Partnership

Social Media Research data

Partnership Overview

Amazon, through the AGI Autonomy Lab, funded a research collaboration with The Ohio State University's Natural Language Processing (NLP) Group (led by PI Professor Huan Sun) to develop state-of-the-art open deep research agents. The goal: build AI systems capable of autonomously browsing the web, synthesizing information, and delivering comprehensive, citation-backed answers — and make the results freely available to the global research community.

This partnership represents a strategic alignment between Amazon's frontier AI research capabilities and Ohio State's world-class NLP expertise, producing results that have captured global attention. This research has applications for internal research & development efforts as well, across Amazon.

Investment Component & Details: Direct Research Funding was $260,000 cash to OSU NLP Group. Technical Collaboration: Amazon AGI Autonomy Lab provided guidance through bi-weekly meetings.

Project: QUEST Deep Research Agent

Research Objective

Train a state-of-the-art open deep research agent based on open-weight LLMs that can handle diverse search and research tasks — from fact-seeking questions to long-form report synthesis with citations — and publicly share all training details, data, and code to advance the field.

Technical Innovation

  • Rubric-Tree Data Synthesis: Novel pipeline that generates training tasks with structured, verifiable evaluation criteria — no human annotation required
  • Three-Stage Training Pipeline: Mid-training (context management), supervised fine-tuning (trajectory learning), and reinforcement learning (reward-based optimization)
  • Context Management System: Built-in mechanism that enables effective long-horizon reasoning, compressing interaction history into structured states for coherent multi-step research
  • Scalable Model Family: Models ranging from 2B to 35B parameters, enabling deployment from lightweight local agents to full-scale research systems

Results & Impact

Benchmark Performance

QUEST was evaluated across eight deep research benchmarks spanning fact-seeking, citation grounding, and report synthesis tasks. Key results:

  • Approaches or surpasses frontier closed-source agents (including proprietary systems from OpenAI and Anthropic) using only 8,000 synthesized training tasks
  • Best overall performance among open-weight agents across all evaluated benchmarks
  • Even the smallest model (QUEST-2B) achieves strong fact-seeking scores (30.3 on HLE, 72.8 on GAIA), demonstrating the recipe's effectiveness
  • Paper jointly submitted to NeurIPS 2026 — a top-tier machine learning conference

Open-Source Deliverables

In the spirit of open science, the team released the complete package:

  • Model Weights: Full family (2B, 4B, 9B, 35B) on Hugging Face
  • Training Data & Scripts: Complete synthetic data pipeline and training framework
  • Research Paper: Full paper jointly published on arXiv (arXiv:2605.24218)
  • Live Demo: Interactive Hugging Face Space for public experimentation
  • Source Code: Complete codebase on GitHub

Public Recognition & Amazon Attribution

The research announcement on X (formerly Twitter) by the OSU NLP Group went viral within hours, accumulating 2.5+ million views overnight. Critically, Amazon is acknowledged as the funding

partner and co-author in both the X post and the published paper, providing significant brand visibility in the AI research community.

The Amazon AGI Autonomy Lab is also planning to integrate QUEST's data and training recipe into their general agent training production efforts, further extending the impact of this collaboration.

Published Links

Strategic Value

This partnership delivers outsized value on multiple dimensions:

  • Global Research Leadership: OSU's NLP Group has produced an open-source AI agent that competes with the world's leading AI labs (OpenAI, Anthropic). This positions Ohio State as a top-tier institution in frontier AI research.
  • Democratizing AI: By fully open-sourcing models, data, and training scripts, OSU and Amazon are advancing the mission of making powerful AI accessible to all researchers and developers worldwide.
  • Proven Industry Partnership Model: This collaboration demonstrates how university-industry partnerships can produce world-class results in a compressed timeframe, with Amazon providing funding and infrastructure while OSU delivers research excellence.
  • Massive Visibility: The viral reception (2.5M+ views overnight) generates significant attention for both OSU and Amazon, attracting top research talent and future collaboration opportunities.
  • Ongoing Amazon Integration: Amazon's AGI Autonomy Lab is integrating QUEST's training data and recipe into production agent systems, ensuring continued collaboration and future research phases.

Prepared by Steve Fisk, Enterprise Account Manager — Higher Education, AWS
June 18, 2026