US AI Labs Cut Prices 25% in 1 Month to Fend Off Chinese Rivals

OpenAI's July 30 price list made the split explicit. GPT-5.6 Luna dropped 80% and GPT-5.6 Terra fell 20%. GPT-5.6 Sol, the flagship frontier model, got faster API performance at an unchanged price. The company said in a blog post that it was improving performance per dollar across enterprise workloads. It described the work as building "a resilient infrastructure portfolio" that matches each workload to the system best suited to run it, supporting "both ends of the price-performance curve."

Anthropic's launch followed the same shape: Claude Opus 5 entered at half the price of its predecessor, Fable 5, and the report grouped the move with OpenAI's mid-tier reductions. Neither lab cut its flagship line.

The moves land two months into a price war that has reached U.S. consumers and enterprise procurement alike. Soaring AI costs pushed some enterprise buyers toward the cheaper Chinese models, and the era of "tokenmaxxing," the practice of pushing employees onto the biggest models with the heaviest usage, is ending after two years of unchecked growth.

DeepSeek is answering on price. The Chinese lab is adding peak-hour pricing for its flagship V4 models that will quadruple current levels, according to a report Thursday, though the new rates remain below what its main competitors charge.