OpenAI says its new GPT 5.6 models are becoming more cost-efficient

OpenAI says it has cut the price of two GPT-5.6 models, Luna and Terra, as part of a push to make those models more cost-efficient. Luna’s API price is down 80 percent. Terra’s API price is down 20 percent. The cuts apply to API access for those two models, not to a broader product line beyond what was reported.

On the product side, the move is framed as efficiency work rather than a rebrand or a new capability drop. OpenAI is pricing Luna and Terra lower while stating that the models themselves are becoming more efficient. The only concrete mechanics disclosed are the two API price changes and the size of each cut: an 80 percent reduction for Luna and a 20 percent reduction for Terra under the GPT-5.6 naming.

For engineers and builders on the API, those numbers change the unit economics of calling GPT-5.6. An 80 percent drop on Luna is a large shift in cost per request for workloads that can use that model. A 20 percent drop on Terra is smaller but still material for high-volume or always-on usage. Teams comparing models on quality versus spend will re-run cost models with these two prices as the new baseline.

In market terms, OpenAI is using price and efficiency claims together on named GPT-5.6 models rather than holding API rates fixed while only talking about performance. Cutting Luna far harder than Terra signals different positioning or cost structure between the two, even without further detail on how each is built. Other providers compete on the same axis of model quality and API cost; these reductions move OpenAI’s offer on that axis for Luna and Terra specifically.

What to watch next is whether the lower API prices hold as usage grows, and whether OpenAI extends the same efficiency-and-price pattern beyond Luna and Terra. Builders should recheck Luna and Terra in routing and budget plans now that Luna is 80 percent cheaper and Terra is 20 percent cheaper on the API, and track whether further GPT-5.6 pricing or efficiency claims follow.