Jakub Pachocki is OpenAI's Chief Scientist, as named in a September 2025 OpenAI developer blog post, and is listed with an OpenAI affiliation on a September 2026 paper. The GPT-4 Technical Report credits Pachocki as overall lead and optimization lead for pretraining, and the Dota 2 paper says Pachocki and Szymon Sidor set research direction throughout the OpenAI Five project. The OpenAI o1 System Card lists Pachocki in both the reasoning research leadership and executive leadership, and Pachocki appears among the contributors to the gpt-oss-120b and gpt-oss-20b model card.123456
At a glance
- Current role
- Chief Scientist
- Company
- OpenAI
- Sector
- Company
- Work listed
- 4
- Sources cited
- 15
- Profile reviewed
Algorithms research
Pachocki's early papers were in theoretical computer science, on graph algorithms and numerical linear algebra. "Geometric Median in Nearly Linear Time" (2016), with Michael B. Cohen, Yin Tat Lee, Gary Miller and Aaron Sidford, gave faster algorithms for a classic optimization problem. The same year Pachocki posted the single-author "Analysis of Resparsification" on matrix sparsification and co-authored a paper on fast algorithms for approximating girth and roundtrip spanners in directed graphs.789
Research at OpenAI
At OpenAI Pachocki worked on reinforcement learning. Pachocki co-authored "Emergent Complexity via Multi-Agent Competition" (2017) and the 2018 paper on learning dexterous in-hand manipulation with a robotic hand, and the Dota 2 paper credits Pachocki and Szymon Sidor with setting research direction throughout the OpenAI Five project.10114
Pachocki later worked on large language models. Pachocki co-authored "Tensor Programs V" (2022) on transferring hyperparameters from small to large networks, and the GPT-4 Technical Report credits Pachocki as overall lead and optimization lead for pretraining. The OpenAI o1 System Card lists Pachocki in reasoning research leadership and executive leadership, and Pachocki co-authored the 2025 paper on competitive programming with large reasoning models.123513
As OpenAI's Chief Scientist, Pachocki is among the contributors to the gpt-oss-120b and gpt-oss-20b model card, whose models are released under the Apache License 2.0. Pachocki also co-authored the 2025 position paper "Chain of Thought Monitorability", which discusses monitoring models' reasoning as an opportunity for AI safety.161415
Work and contributions
gpt-oss Open weights
Listed among the contributors to the gpt-oss-120b and gpt-oss-20b model card (2025); the openai/gpt-oss repository is under the Apache 2.0 license.614
GPT-4 pretraining Proprietary product
Credited in the GPT-4 Technical Report (2023) as overall lead and optimization lead among the pretraining core contributors.3
OpenAI Five Published research
Co-author of "Dota 2 with Large Scale Deep Reinforcement Learning" (2019); the paper's contributions section says Pachocki and Szymon Sidor set research direction throughout the project, including the first version of the Rapid system for large-scale RL.4
OpenAI o1 Proprietary product
Listed under reasoning research leadership and executive leadership in the OpenAI o1 System Card (2024).5
Timeline
- Posted "Analysis of Resparsification" as sole author.8
- Co-authored "Geometric Median in Nearly Linear Time".7
- Co-authored "Emergent Complexity via Multi-Agent Competition".10
- Co-authored "Learning Dexterous In-Hand Manipulation".11
- Co-authored the OpenAI Five Dota 2 paper, credited with setting research direction.4
- Co-authored "Tensor Programs V" on zero-shot hyperparameter transfer.12
- Credited as overall pretraining lead in the GPT-4 Technical Report.3
- Listed in reasoning research and executive leadership in the OpenAI o1 System Card.5
- Co-authored "Competitive Programming with Large Reasoning Models".13
- Contributed to the gpt-oss-120b and gpt-oss-20b open-weight model card.6
Where they work
- OpenAI 63 people in the wiki
Related people
- Susan ZhangAlso at OpenAI
- Eric MitchellAlso at OpenAI
- Noam BrownAlso at OpenAI
- Aravind SrinivasAlso at OpenAI
- Hyung Won ChungAlso at OpenAI
- Weizhu ChenLanguage models · Training systems
- Johannes HagemannTraining systems · Reinforcement learning
- Sami JaghouarTraining systems · Reinforcement learning
- Edward BeechingLanguage models · Reinforcement learning
- Mika SenghaasReinforcement learning · Training systems
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
15 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026
- Why we built the Responses API (OpenAI Developers blog)
- What if automating AI R&D triggers an intelligence explosion? (arXiv 2609.36054), author affiliations
- GPT-4 Technical Report (arXiv 2303.08774), authorship and credit attribution
- Dota 2 with Large Scale Deep Reinforcement Learning (arXiv 1912.06680)
- OpenAI o1 System Card (arXiv 2412.16720), authorship and credit attribution
- gpt-oss-120b & gpt-oss-20b Model Card (arXiv 2508.10925), contributors
- Geometric Median in Nearly Linear Time (arXiv 1606.05225)
- Analysis of Resparsification (arXiv 1605.08194)
- Approximating Cycles in Directed Graphs: Fast Algorithms for Girth and Roundtrip Spanners (arXiv 1611.00721)
- Emergent Complexity via Multi-Agent Competition (arXiv 1710.03748)
- Learning Dexterous In-Hand Manipulation (arXiv 1808.00177)
- Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer (arXiv 2203.03466)
- Competitive Programming with Large Reasoning Models (arXiv 2502.06807)
- openai/gpt-oss repository (Apache License 2.0)
- Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety (arXiv 2507.11473)