OpenAI Releases GPT-6.1 Sol at a Fifth of GPT-6 Astra's Token Prices

My take: OpenAI launched GPT-6.1 Sol with the numbers every team running agents wants to see: near-top performance at one-fifth the cost of its most capable model (Astra). On Terminal-Bench Science, the cost per task drops from $23 to $5.47. That's not a marginal improvement; it's the kind of shift that makes automating previously cost-prohibitive workflows actually viable.

Worth noting: these benchmarks were published by OpenAI itself, which is both the judge and the party being measured. The numbers are a useful reference, but worth reading critically and waiting for independent third-party validation before treating them as definitive.

What each announcement like this makes increasingly clear is that the cost of accessing powerful AI models keeps dropping. That matters: AI is no longer expensive to access; the barrier now is knowing how to use it well. Are you measuring cost per task in your workflows, or are you still choosing models based on abstract rankings alone?