A theory about OpenAI, Anthropic and compute
I have a theory about what's happening with openai and anthropic.
Anthropic was compute constrained for so long because everyone was using their models for work and coding.
This gave openai lots of compute headroom to develop, research, and hand out that compute to their users.
Once openai grabbed such a major market share with better models and usage, they become the ones who were compute constrained.
This gave anthropic the wiggle room they needed to research and develop Opus 5.5 with crazy limits.
Now I'm thinking we might see the pendulum swing again in a month or so.
Or I could be blowing hot air, who knows?
Might be coming back to this tweet in a month.