OpenAI推出了GPT-5.6系列模型的价格调整方案 [1]。其中，入门款Luna大幅降价80%，输入价格从1美元/百万Token降至0.2美元/百万Token，输出价格从6美元/百万Token降至1.2美元/百万Token [1]。中端型号Terra降价20%，输入和输出价格分别为2美元/百万Token和12美元/百万Token [1]。旗舰模型Sol保持原价不变，输入价格为5美元/百万Token、输出价格为30美元/百万Token，但新增了Fast mode加速功能，速度相比标准模式提升2.5倍，其使用价格为标准模式的2倍 [1]。Luna的降价幅度使其输入价格已与上一代GPT-5.4 nano持平 [1]。
降价源于OpenAI在生产系统优化方面的成果 [1]。通过GPU Kernel优化和改进推测解码等措施，公司使端到端服务成本下降了20%，同时Token生成效率提升了15%以上 [1]。OpenAI强调整个优化过程仍然由人类主导，模型虽然参与其中但需要保持"人类在环"的监督机制 [1]。
OpenAI has unveiled a pricing restructuring for its GPT-5.6 model lineup, with the entry-level Luna variant receiving an 80 percent reduction in costs [1]. The Luna model's input price has dropped from $1 to $0.2 per million tokens, while output pricing has fallen from $6 to $1.2 per million tokens [1]. The mid-tier Terra model receives a 20 percent price cut, with input and output costs now set at $2 and $12 per million tokens respectively [1]. In contrast, the flagship Sol model maintains its existing pricing of $5 for input and $30 for output per million tokens, though it now includes a new Fast mode feature that delivers 2.5 times faster performance at double the standard pricing rate [1].
The cost reductions stem from OpenAI's optimization of its production infrastructure, achieved through GPU kernel enhancements and improvements to speculative decoding that have reduced end-to-end service costs by 20 percent and increased token generation efficiency by over 15 percent [1]. OpenAI emphasized that the entire optimization process remains human-guided, with the model participating in improvements while maintaining "Human in the Loop" oversight [1]. The Luna model's pricing now aligns with the input cost of the previous-generation GPT-5.4 nano variant [1].
