According to the analysis of OpenRouter data by the external media, the recent downward revision of part model prices by OpenAi has resulted in a rapid expansion of the developer ' s call and an increase that exceeds the price decrease, and in the short term, the related model ' s revenues have risen. This provides a new sample of observations for the pricing strategy of the AI industry.
The two models have been reduced and used significantly.
OpenAI recently reduced the GPT-5.6 Luna price by 80 per cent and the intermediate model Terra by 20 per cent. About two weeks after the analyst tracked the price adjustment, it was found that the real use price of Luna had dropped to about one tenth and the call volume had increased by about 14 times; the real use price of Terra had dropped to about one third and the use had increased by about 5 times.
Short-term income did not decline
TD Cowen estimates that Luna ' s income increased by about 34 per cent and Terra ' s income by about 45 per cent compared to the seven days before the price reduction. Of these, the change in Luna is of the greatest concern because income growth is not common when prices are reduced by 80 per cent at a time.
The article mentions that the use of the AI model is usually charged token. For business clients, when model prices fall, tasks that are already costly become easier to land, and lead to more high-frequency calls.
Lower prices bring more scenes.
According to the external review, this is in line with a pattern common in the technological industry: when some basic capacity becomes cheaper, the market tends not to reduce its use, but to put it into more new scenes. For AI, enterprises can use lower-cost models for more document processing, client questions and answers, code writing and multi-step automated proxy assignments.
The article also mentioned that Enterprise Expenditure Data, Ramp, had previously found that OpenAI ' s GPT-5.6 Sol had received a larger share of business expenditure than Anthropic ' s Fable 5 in July. The report suggests that lower prices may be an important reason, as lower costs usually lead to higher frequency of use.
The observation period is still short.
The review also notes that the bottom generation costs of AI token are also declining. As the new generation of computing systems improves efficiency, model output costs are likely to continue to decline, which means that future AI pricing and further exploration of space.
However, the time window for this observation was only about two weeks, and the sample period remained short. Analysts also indicated that more time was needed to determine whether the “price-for-growth” round was sustainable.