Andrew Tulloch
Andrew Tulloch is an AI researcher whose work links large-scale machine-learning systems and model training. He grew up in Perth, Australia, studied mathematics at the University of Sydney and Cambridge, and won Sydney’s University Medal for his honours thesis in 2011. At Facebook and Meta, his co-authored papers cover rapid ImageNet training and large-scale recommendation systems. He joined OpenAI in 2023 and helped train GPT-4o, GPT-4.5, and o3 before co-founding Thinking Machines Lab in 2025. As of October 2026, he is at Anthropic.
Timeline
-
2011
A University Medal
Raised in Perth, Australia, he studies mathematics at the University of Sydney and University of Cambridge, and wins Sydney’s University Medal for his honours thesis in 2011.
-
2012
Joining Facebook
He moves to the United States and joins Meta, then known as Facebook, in 2012.
-
2017
Training ImageNet in an hour
With a Caffe2 system, he co-authors a 2017 paper that trains ResNet-50 on 256 GPUs in one hour with minibatches of 8,192.
-
2021
Scaling recommendation systems
A 2021 Facebook paper he co-authors introduces ZionEX, a software-hardware system for distributed training of large recommendation models.
-
2023–2025
Training OpenAI models
During his OpenAI work, he helps train GPT-4o, GPT-4.5, and o3 after joining in 2023.
-
2025
Co-founding Thinking Machines Lab
As a co-founder of Thinking Machines Lab in 2025, he works on pretraining and reasoning.
-
October 2025
Moving to Meta
In October 2025, AI researcher Andrew Tulloch leaves Thinking Machines Lab for Meta.
-
September 2026
Joining Anthropic
At Anthropic, he joins its inference and performance team in September 2026.
Sources and credits
Photo: We have no published photo of Andrew Tulloch yet. Send us a free-licensed photo.
- Andrew Tulloch (B. Science – Hons) – Engineer, Facebook | Fresh Out of Sydney Uni
- Andrew Tulloch
- Meta offered this Australian $1.5 billion. Now he’s joining Anthropic.
- Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
- Software-Hardware Co-design for Fast and Scalable Training of Deep Learning Recommendation Models
- Mira Murati’s Stealth AI Lab Launches Its First Product
- Thinking Machines Lab co-founder Andrew Tulloch heads to Meta