OpenAI Cuts GPT-5.6 Luna Prices by 80% to Fight Chinese Open-Source Models

On July 31, 2026, OpenAI announced a major price cut for its GPT-5.6 model family. The light-tier model, called Luna, had both its input and output prices reduced by 80%. After the cut, one million input tokens cost only 20 cents, while one million output tokens cost about 1 dollar and 20 cents. The mid-tier model, Terra, was also cut, though by a smaller 20%. The top-tier flagship model, Sol, kept its original price. Why cut prices so sharply? The answer is competition. On one major AI model platform, Chinese open-source models now hold about 63% of global market share, while US models sit at around 35%. Low-cost Chinese models, including DeepSeek V4 and Kimi K3, have been winning customers with strong performance and very small bills. To stay in the game, OpenAI is now matching or beating those prices on its lighter models. The move marks a clear turning point in AI business strategy. For years, the industry competed mainly on raw capability, training ever-larger models on more chips. In 2026, the race is shifting toward who can offer the best price for useful intelligence. When a frontier company like OpenAI cuts prices by 80% on its own, it sends a strong signal to the whole market: the era of "premium-only AI" is ending. For customers, the impact is immediate and very practical. A startup that once spent 10,000 dollars a month on AI calls could now spend only 2,000 dollars for the same volume. That means smaller teams can run chatbots, search tools, and content pipelines that were once too expensive. The question everyone is asking now is simple: with prices falling this fast, can AI companies still earn enough to keep building better models? The answer will shape the next decade of the industry.