Is GPT-6.1 Sol just Astra on a budget?

It lands within a whisker of the flagship on plenty of jobs at a fifth of the price. So what is the flagship even for?

Quick lineup check. OpenAI's GPT-6 family has tiers, with Astra as the top dog and Sol as the faster, cheaper one. GPT-6.1 Sol is not a bigger model. It is a better Sol, and it shipped a week after the first one.

Here is where it gets fun. On the office work and coding style tests OpenAI ran, the new Sol lands close to Astra. On the computer use benchmark it sits about two points under, for roughly a seventh of the cost per task. On DeepSWE, a hard test built around real software projects, it scores 75.2%, ahead of the best Astra result. And the price? $2 per million input tokens and $10 for output, against $10 and $50 for Astra.

Hold on. If the cheap one is that close, why does Astra exist at all?

Because the gap has not vanished, it has moved. On a science heavy test called Terminal-Bench Science, the new Sol scores 57.0% while Astra hits 68.1%. The cost per task there is $5.47 against $23.80. So you pay more than four times as much for eleven extra points. Whether that is worth it depends entirely on how hard your problems are.

One more thing worth knowing: most of these numbers were reported by OpenAI itself. Treat them as a strong hint, not a verdict.