OpenAI Just Made Frontier AI 80% Cheaper to Run

On July 30, OpenAI cut the price of its GPT-5.6 Luna model by 80%, down to 20 cents per million input tokens and $1.20 per million output tokens. Terra, the mid-tier model, dropped 20%. Three weeks later, on August 21, OpenAI cut pricing again, this time on its top-tier Sol model, slashing input costs 20% to $4 and output costs by a third to $20 per million tokens for the next three months.

For anyone outside the industry, the token prices don't mean much on their own. What matters is what they unlock: running a capable AI model used to cost enough that only well-funded companies could afford to use it at scale. At these new prices, a small startup or even a solo developer can run serious workloads, automation, customer support, document processing, without the bill wiping out the budget.

This isn't happening in isolation. Rivals are moving the same direction: Alibaba pushed out its Qwen 3.8-Max model positioning it for software development work, and a former xAI co-founder raised $1.1 billion for River AI, a startup built specifically around open-source models people can run and control themselves instead of renting access from a big lab. Anthropic, meanwhile, has been pushing in the other direction on capability, expanding context windows to a million tokens so models can work with much larger documents at once.

Cheaper AI access doesn't guarantee it gets used well. But it does lower the barrier for who gets to build with it, and that's the part worth watching: a year ago, this level of capability was priced for enterprises. Now it's priced for anyone with an idea and a laptop.