OpenAI finishes Spud pretraining, Altman shifts focus

Potato jokes aside, this looks like OpenAI clearing the runway for the next frontier-model cycle. The real signal is the operating model shift: Altman closer to the technical loop, Lightcap closer to the machinery that ships and scales it.

• –Finished pretraining usually means the compute-heavy phase is over, but the model still has to prove itself in post-training, evals, and safety work.
• –If Spud is a frontier base model, OpenAI is still betting that raw pretraining can buy meaningful headroom.
• –A tighter research-to-product loop tends to shorten the path from internal breakthrough to API release, which is what developers actually feel.
• –Faster cadence is a double-edged sword: better models sooner, but more churn in behavior, compatibility, and pricing expectations.