OpenAI just shipped a model that scores lower.

On two of six headline benchmarks, GPT-6 Sol lands under the model it replaces. It also costs about 60% less per task.

That wasn't a miss. That was the pitch.

Anthropic launched Claude Opus 5.5 hours earlier. Twenty percent cheaper per token, about 40% cheaper on real workloads because it needs fewer tokens to do the same job.

Two frontier labs, same morning, and neither raised the ceiling.

Ten days ago Anthropic's CEO published an essay arguing the industry must slow the rate at which it improves capabilities.

You could read today as the rebuttal. Two launches, ten days later.

But look at what actually shipped. Nobody pushed the frontier. Both competed on price.

I wrote in July that AI was leaving its horsepower era and entering its mileage era. Cars made that turn when gas got expensive. It took Detroit decades. This took ten weeks.

Here's what I can't resolve.

I don't know whether this is restraint, or whether cheap open-weight models made capability a bad place to compete.

Probably both. The incentive and the ethics pointed the same way this month.

We learn nothing about which one was driving until they point in opposite directions.

What should your model provider optimize for next year, smarter or cheaper?