OpenAI VP and veteran researcher Jerry Tworek steps down, here's why

OpenAI

• Jerry Tworek leaves OpenAI after 7 years to explore new research opportunities
• Tworek developed o1 reasoning model, contributed to Codex, Copilot, GPT-4
• His departure is a significant loss for OpenAI's advanced reasoning systems shift.

Did our AI summary help?

Jerry Tworek announced his resignation from OpenAI after nearly seven years, confirming the move in a post on X where he shared an internal memo sent to his team. He said the decision was driven by a desire to explore research paths that are increasingly difficult to pursue within OpenAI’s present priorities.

While acknowledging the emotional weight of the move, Tworek described OpenAI as an institution with a permanent place in history, arguing that its contributions to large-scale reasoning systems have already had a lasting impact on humanity.

A central figure in OpenAI’s reasoning shift

Tworek joined OpenAI in 2019, when it still largely operated as a nonprofit research lab. Over time, he became a key architect behind its push toward advanced reasoning models and practical AI systems for developers.

His most visible contribution was leading the development of the o1 reasoning model, which marked a shift away from pure pretraining toward reinforcement learning-based reasoning. Instead of relying only on pattern recognition, o1 was designed to reason through problems step by step, evaluating intermediate decisions along the way. The approach paid off, with o1 topping LiveBench’s leaderboard in August for complex reasoning tasks.

From Codex to Copilot and GPT-4

Beyond reasoning models, Tworek played a significant role in OpenAI’s developer ecosystem. He led research on Codex, which later became the foundation for GitHub Copilot, and contributed directly to the coding capabilities embedded in GPT-4 and ChatGPT.

In his farewell note, he reflected on scaling reinforcement learning on robots, training early coding models, and helping uncover the chinchilla scaling laws that reshaped how model size and data efficiency are balanced. He also pointed to team-building as one of his proudest achievements, noting that many of the groups he helped create went on to define OpenAI’s modern reasoning paradigm.

Tworek described his time at OpenAI as deeply formative, both personally and professionally. He spoke about long nights, close friendships, and the rare chance to witness multiple foundational AI breakthroughs from inside the organisation.

While departures are never easy, he framed the moment as an opening for new ideas and renewed progress. His exit marks the loss of a researcher closely tied to OpenAI’s transformation from an experimental lab into one of the most influential forces shaping modern AI reasoning and developer tools.