View: OpenAI and Anthropic May Keep Strongest Models Internal

haider1 · x · 2026-08-21

The author suggests that OpenAI and Anthropic are pushing hard internally. As they approach recursive self-improvement, it makes less sense to give competitors the best model.

The argument is that shipping a model 1-2 generations behind is still sufficient to be SOTA, while keeping the strongest model internal allows them to stay ahead in research.

Related event: Report: OpenAI and Anthropic May Be Holding Back Best Models for IPO(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →