More Powerful AI Is Finally Costing Less

More State Change insights on using AI in business: https://statechange.ai This is a great time to be on the customer side of AI. We're getting more power and more value per token, while competition puts pressure on the premium the frontier labs can charge. I look at what the old frontier cost, today's lower-priced choices, and the open-weight models chasing the frontier from China and beyond. Then I ask what that means for your spending plan—and whether two very good models checking each other's work could be cheaper than one premium model. Reconsider the assumption that more powerful models have to be more expensive models. The charts compare USD per million output tokens, using prices observed October 16–17, 2026. These are selected models, not equal-performance choices or a price cut to every model. Cheaper tokens do not guarantee a cheaper successful task: token use and task success still matter. Evidence notes Artificial Analysis scores are a dated composite-intelligence snapshot, not proof of a historical coding catch-up trend. Closed-model parameter sizes are undisclosed. The historical $180 anchor is GPT-5.5 Pro; Fable 5.1 was introduced September 1, 2026. Open-weight licenses and availability vary, and developer geography does not determine hosting location. Benchmark reference: https://artificialanalysis.ai/ Chapters 00:00 More power, lower prices 00:34 What the old frontier cost 02:22 What the current frontier costs 03:59 Chinese open weights and the price mechanism 07:11 Western open-weight counterparts 08:45 What this means for you