The Middle Market Is Shaping AI Economics

Summary

AI model prices are falling rapidly, but the fiercest competition is centered on the middle of the market rather than the frontier. The article argues that most business use involves multi-step workflows that need sufficient intelligence at a manageable cost. Anthropic kept its Opus models at $5 and $25 per million tokens across five releases before making its first cut, while OpenAI reduced Luna’s price by 80% in July and another 50% in September. OpenAI also priced GPT-5.6 Sol at $5 and $30 per million tokens after Anthropic had set frontier pricing at $10 and $50 with Fable 5. Open models add further pressure: they account for a majority of token volume on gateways that publish data and cost 86% less than closed models on a blended basis. Fine-tuned open-weight systems produced additional savings for Cursor and Harvey, with reported cost reductions of 86% and 55%, respectively. At the expensive end, Anthropic’s Fable 5.1 represented only 3.7% of gateway spending during its first twelve days, while corporate frontier-model consumption fell from 53% to 45% between early August and September. The article concludes that demand resembles a fat middle rather than a pyramid with a highly profitable peak. As intelligence per dollar improves, the models that meet relatively stable enterprise requirements may become increasingly commoditized, shaping the economics of the AI market.