Prafulla Dhariwal

Prafulla Dhariwal’s career traces an unusually broad arc through modern generative modelling: reinforcement learning, flows, music, diffusion, language and native image generation. His recent leadership has helped make multimodal creation a core model capability rather than a separate specialist pipeline.

Prafulla Dhariwal is an OpenAI research scientist and, in the company’s December 2025 project credits, its multimodal lead. Raised in Pune and educated at MIT, he has contributed to several distinct lines of modern AI research. He co-authored Proximal Policy Optimisation, introduced Glow with Diederik Kingma, led Jukebox, co-authored GPT-3, and was an equal-first author of work behind DALL·E 2 and diffusion models that surpassed GAN baselines in image synthesis. The recent record brings those strands into general-purpose multimodal systems. OpenAI’s March 2025 native image-generation release names Dhariwal under Multimodal Organization leadership and foundational research. Unlike a detached image tool, the system generated and edited images within the same multimodal model used for conversation. In December, the GPT Image 1.5 contribution record named him Multimodal Lead. The company reported more reliable instruction following and detail-preserving edits, generation up to four times faster, and image input and output priced 20% below GPT Image 1; these are company-reported product comparisons. Dhariwal’s case is not that he built these systems alone. It is the depth of his individual research record combined with documented leadership of teams translating generative methods into widely available multimodal models.