The Price of AI Capability Is Falling Fast

A new OpenAI model, GPT 6.1 Terra, has been released.

It does not match GPT-6 Astra in strength, but it is much more efficient and points to the direction AI is taking. The trend is toward commoditisation, and it is outpacing AI's arrival as an essential service.

The Sol model was already close to Astra at one-fifth of the price, and Terra now halves that cost again. OpenAI also says the model is especially suited to automated computer use.

"Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it."

I am dubious about the 'writes naturally' claim from any model, yet steady gains are visible across every task. When it comes to markets, lower prices make adoption more attractive, but they also raise questions about margins and returns on capital for companies supplying AI infrastructure.

Ultimately, I firmly believe the gains from AI will go to established, asset-heavy companies that deploy it to boost efficiency, not to the model builders. The bulls will point to Jevon's Paradox, and that argument could work for some time, measured in years, yet models are becoming so capable and inexpensive that before long there will be ultra-cheap versions covering 99% of use cases.

On that path, profits move from tech companies to productivity beneficiaries, with job cuts occurring along the way. Consider a firm with large labour costs and little exposure to AI-related capital spending. Automating 10–20% of everyday office tasks could have a massive impact on its profit margins.

Most recently, the Nasdaq was up 1.0%, and the day's big news is that Anthropic has filed for an IPO.