AI Just Got a Lot Cheaper. Here's Why That Actually Matters.

On 29 September 2026, OpenAI released GPT-6.1 Sol: a model it describes as near-flagship intelligence for coding, computer use and professional work, priced at a fifth of its top-tier GPT-6 Astra model. That's not a minor pricing tweak. It's the kind of shift that changes who AI is actually affordable for.

The numbers, plainly

GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens, against the $10 and $50 Astra charges for the same. In practice, that's the difference between a tool priced for enterprise budgets and one priced for a business running lean.

This is part of a pattern, not a one-off

OpenAI isn't moving alone. Other labs have released comparably priced frontier models in the same window, part of a broader trend this year: genuinely capable AI getting meaningfully cheaper every few months, not just at one company. The cost of trying this properly has been dropping all year, and this is the most visible step yet.

Why this matters if you've been waiting

Cost has been a real, legitimate reason to wait on experimenting with AI, not just a convenient excuse. When the leading models cost enterprise money to run at any real volume, "wait and see" was a sensible default for a small or medium business. A fifth of the price doesn't make every use case suddenly worth it, but it does mean the maths on a well-defined, repetitive task now works out differently than it did even two months ago.

What this doesn't mean

It doesn't mean you need to implement AI across the business tomorrow. Price was one barrier among several, alongside skills and knowing which process to actually point it at, and this only moves one of them. But if cost was specifically the reason you were waiting, that reason just got five times weaker.