Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

OpenThoughts-Agent: Data Recipes for Agentic Models

Source

arXiv

arxiv.org/abs/2606.24855v1

Read the source

Source

  • Source name: arXiv
  • Source kind: paper
  • Authors / organization: Negin Raoof, Richard Zhuang, Marianna Nezhurina, Etash Guha, Atula Tejaswi, Ryan Marten, Charlie F. Ruan, Tyler Griggs, Alexander Glenn Shaw, Hritik Bansal, E. Kelly Buchanan, Artem Gazizov, Reinhard Heckel, Chinmay Hegde, Sankalp Jajee, Daanish Khazi, Emmanouil Koukoumidis, Xiangyi Li, Hange Liu, Shlok Natarajan, Harsh Raj, Nicholas Roberts, Ethan Shen, Nishad Singhi, Michael Siu, Ashima Suvarna, Hanwen Xing, Patrick Yubeaton, Robert Zhang, Leon Liangyu Chen, Xiaokun Chen, Steven Dillmann, Saadia Gabriel, Xunyi Jiang, Anurag Kashyap, Boxuan Li, Yein Park, Minh Pham, Sujay Sanghavi, Lin Shi, Ke Sun, Yixin Wang, Zhiwei Xu, Erica Zhang, Siyan Zhao, Wanjia Zhao, Jenia Jitsev, Alex Dimakis, Benjamin Feuer, Ludwig Schmidt
  • Venue: arXiv: cs.AI
  • Published: 2026-06-23
  • Collected: 2026-06-24
  • URL: http://arxiv.org/abs/2606.24855v1
  • PDF: https://arxiv.org/pdf/2606.24855v1
  • DOI: UNKNOWN

Plain-Language Brief

The OpenThoughts-Agent project presents a fully open data curation pipeline for training broadly capable AI agents. The researchers conducted ablation experiments and used the pipeline to create a dataset of 100K examples, which they used to fine-tune Qwen3-32B. This resulted in an improvement of 3.9 percentage points over the strongest existing open data agentic model (Nemotron-Terminal-32B). The researchers also found that their training data exhibits strong scaling properties.

What the Document Says

  • The OpenThoughts-Agent (OT-Agent) project presents a data curation pipeline for training broadly capable AI agents.
  • The authors conducted over 100 controlled ablation experiments to investigate the pipeline.
  • They created a training set of 100K examples using the pipeline and fine-tuned Qwen3-32B on this dataset.
  • Fine-tuning Qwen3-32B resulted in an average accuracy of 44.8% across seven agentic benchmarks.
  • This is a 3.9 percentage point improvement over the strongest existing open data agentic model, Nemotron-Terminal-32B (40.9%).
  • The training data exhibits strong scaling properties.
  • The authors publicly release their training sets, data pipeline, experimental data, and models at openthoughts.ai.

Key Concepts Mentioned

  • OpenThoughts-Agent
  • data curation pipeline
  • broadly capable AI agents
  • ablation experiments
  • Qwen3-32B
  • training set
  • 7 agentic benchmarks
  • Nemotron-Terminal-32B
  • scaling properties

Claims Made by the Document

  • The OpenThoughts-Agent project presents a data curation pipeline for training broadly capable AI agents.
  • Fine-tuning Qwen3-32B on the OT-Agent training set resulted in an improvement of 3.9 percentage points over Nemotron-Terminal-32B.
  • The training data exhibits strong scaling properties.

Evidence Used

  • The authors describe the OpenThoughts-Agent project and its purpose.
  • The authors state that fine-tuning Qwen3-32B on their training set resulted in an average accuracy of 44.8% across seven benchmarks, with a 3.9 percentage point improvement over Nemotron-Terminal-32B.
  • The authors mention that their training data outperforms alternative open datasets at every training set size in compute-controlled comparisons.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Agentic language models dramatically expand the applications of AI yet little is publicly known about how to curate training data for broadly capable agents. Existing open efforts such as SWE-Smith, SERA, and Nemotron-Terminal typically target a single benchmark, leaving open the question of how to train models that generalize across diverse agentic tasks. The OpenThoughts-Agent (OT-Agent) project addresses this gap with a fully open data curation pipeline for training agentic models. We conduct more than 100 controlled ablation experiments to systematically investigate each stage of the pipeline, yielding insights on the importance of task sources and diversity. We then assemble a training set of 100K examples from our pipeline and fine-tune Qwen3-32B on this dataset, which yields an average accuracy of 44.8% across seven agentic benchmarks and a 3.9 percentage point improvement over the strongest existing open data agentic model (Nemotron-Terminal-32B, 40.9%). Moreover, our training data exhibits strong scaling properties, outperforming alternative open datasets at every training set size in compute-controlled comparisons. We publicly release our training sets, data pipeline, experimental data, and models at openthoughts.ai to support future open research on agentic model training.