OpenAI has reduced the prices of two models in its GPT-5.6 family. Terra, positioned as the balanced option for everyday work, now costs 20 percent less than the rates set earlier this month. Luna, the faster and smaller model, sees an 80 percent cut. The flagship Sol remains unchanged in this round of adjustments.
For developers using the API, the new rates stand at two dollars per million input tokens and twelve dollars per million output tokens for Terra. Luna drops to twenty cents per million input tokens and one dollar twenty cents per million output tokens. These figures mark a clear step down from the initial pricing that accompanied the models’ recent introduction.
Subscribers to ChatGPT Work and Codex do not see changes to their monthly fees. Instead, OpenAI has recalibrated how token consumption is counted against those plans. The practical result is that users can run more queries with Terra and Luna before reaching their usage limits. The company frames the shift as a direct reflection of the lower underlying costs.
Price cuts of this kind have become a recurring pattern in large language models. Providers periodically lower rates as infrastructure costs decline, competition intensifies, and volume grows. Earlier generations of GPT models followed similar trajectories, with successive versions arriving at higher capability and then gradually becoming more accessible. The same dynamic appears here, though the scale of the Luna reduction stands out for how steeply it undercuts the previous entry-level rate.
Whether the adjustments meaningfully expand access depends on the workload. Heavy users of the API will notice the difference most clearly in their monthly bills. Subscription customers gain additional headroom rather than a lower sticker price, which may prove useful for teams already operating near their caps but does little for those who rarely approach the limits. The distinction between raw token pricing and subscription metering continues to shape how different groups experience the same underlying models.
OpenAI’s decision arrives against a backdrop of rapid iteration across the sector. Competing systems from other labs have also seen pricing pressure as providers race to capture developer mindshare and enterprise budgets. In that environment, aggressive cuts on mid-tier and lightweight models serve as one lever among several. They do not alter the fundamental economics of training or inference, yet they do change the near-term cost of experimentation and routine use.
The revised rates for Terra and Luna therefore function less as a dramatic breakthrough and more as a recalibration. They lower the barrier for certain classes of work while leaving the broader structure of OpenAI’s pricing tiers and subscription plans intact. Users will decide for themselves whether the extra capacity or reduced API spend justifies shifting more of their traffic toward these two models.
