AI Just Got 50% Cheaper to Run. Here Is What That Changes.

Yesterday, OpenAI released two new AI models — GPT-6 Sol and GPT-6 Luna — at half the price of the models they replace. For anyone building software that uses AI, that is the kind of change that makes you stop and recalculate. AI just got 50% cheaper to run.

Sol handles complex work: writing and reviewing code, analysing data, working through hard problems. Luna is built for simple, high-volume tasks — summarising documents, answering customer questions, extracting information from long texts. Luna costs $0.10 per million tokens (tokens are the small chunks of text an AI reads and writes — a million of them is roughly a 750,000-word novel). Sol costs $2 per million. Both are exactly half the price of their GPT-5.6 predecessors.

The trade-off is worth knowing: neither new model tops GPT-5.6 on every benchmark. In some coding tests, the older model still scores higher. But OpenAI said these are permanent prices — not introductory rates. For most day-to-day automation tasks, the cost saving is bigger than the performance gap. A business building a document-summarisation tool or a customer support bot does not need the absolute top scorer; it needs a model that is good enough and affordable to run at scale.

If your team has been holding back on an AI feature because the costs did not add up, now is the time to check. A workflow that cost $400 a month on GPT-5.6 now costs around $200 on Luna. That is the difference between a feature that barely breaks even and one that clearly makes sense. Do the maths again — some things that were too expensive last week are not too expensive today.