Suvansh Sanjeev

Working on something new.

I previously led an RL research team at OpenAI, where I co-created algorithms that reshaped OpenAI's post-training from GPT-5 onwards, across modalities. Most recently, I was working on explorations within pretraining. Before all that, I was a contributor on o1 and helped kickstart OpenAI's early robotics revival in 2024, including training the first VLAs at OpenAI. My work was done across teams led by Sébastien Bubeck, Nick Ryder, Shengjia Zhao, and Boris Power.

Before that, I was on leave from my PhD at the Robotics Institute at Carnegie Mellon University, advised by Zico Kolter and Zac Manchester. I studied EECS at UC Berkeley, where I worked with Sergey Levine and Claire Tomlin in the BAIR Lab on deep RL and safe learning.