OpenAI Developing 'Doug', Its Largest Pre-training Model Yet
机器之心 · wechat · 2026-08-09
According to multiple sources, OpenAI is advancing a massive new pre-training model codenamed Doug, which will be its largest to date and is expected no later than November.
The article reviews OpenAI's recent trajectory: since GPT-4o, capability gains relied heavily on post-training and inference-time compute without a generational base model leap. Facing competitive pressure from Google's Gemini 3, OpenAI triggered a 'Code Red'.
Subsequently, OpenAI reportedly resolved pre-training bottlenecks, developing the 'Garlic' model to validate these fixes. Doug represents the scaling of these validated methods. This marks OpenAI's renewed focus on base model scaling after heavily relying on RL and inference-time compute.
Related event: OpenAI Overcomes Pretraining Bottleneck, New Model 'Doug' May Be GPT-6(10 posts)→
More from Companies & People
- US universities go on an AI philosopher hiring spree as philosophy departments add tenure-track AI roles — AnnaCiaunica · 2026-10-16
- NVIDIA and Mercedes-Benz Expand Partnership on Next-Gen Automated Driving — nvidia · 2026-10-16
- Ex-Amazon Engineer: AI Has Killed Bezos's Six-Page Memo Culture — bibryam · 2026-10-16
- DHH's Omarchy Shenzhen Meetup Draws 400+ With Demos on Xiaomi, ARM and iPhone Ports — LoganGrasby · 2026-10-16
- Richard Zach's free 267-page metalogic textbook joins the open textbook collection — burkov · 2026-10-16
- Linus Torvalds slams 'don't use AI' rules: it's like banning compilers for kernel code — G0dZylla · 2026-10-16