OpenAI 于 7 月 31 日宣布下调 GPT-5.6 系列模型的 API 调用费用[1][2]。其中，入门款 Luna 模型降价幅度最大，达到 80%[1][2]，输入价格从 1 美元下降至 0.2 美元，输出价格从 6 美元下降至 1.2 美元[2]。同时发布的 Terra 模型降价 20%，输入和输出价格分别为 2 美元和 12 美元[2]。旗舰款 Sol 模型保持原价，输入价格维持 5 美元、输出价格维持 30 美元，但新增 Fast mode 加速功能，速度提升 2.5 倍，价格为标准模式的 2 倍[2]。
OpenAI 表示此次调价源于生产环节的优化成果[2]。通过优化 GPU Kernel 和推测解码技术，GPT-5.6 Sol 实现了端到端服务成本下降 20% 和 Token 生成效率提升 15% 以上[2]。OpenAI 官方表示"普及先进智能是公司的核心使命，通过不断降低客户使用成本，确保通用人工智能（AGI）造福全人类"[1]。根据该公司预计，ChatGPT 和 Codex 中 Auto-review 模型成本将降至原来的约十分之一[2]。
据 Artificial Analysis Intelligence Index v4.1 的对比，GPT-5.6 Luna 在每完成任务费用方面优于竞品 DeepSeek V4 Pro[1]。
OpenAI announced on July 31 a significant price reduction across its GPT-5.6 model lineup, with the Luna variant experiencing the steepest cut [2]. The Luna model's input price dropped from $1 to $0.2 per unit, while output pricing fell from $6 to $1.2, representing a maximum reduction of 80% [2]. The Terra model received a more modest 20% price cut, with input and output costs set at $2 and $12 respectively [2]. The flagship Sol model maintained its original pricing of $5 for input and $30 for output, but now includes a new Fast mode that achieves 2.5 times faster processing at double the standard rate [2].
OpenAI attributed the price reductions to operational improvements achieved through optimizing the production environment's GPU kernel and implementing speculative decoding technology, which collectively reduced end-to-end service costs by 20% and improved token generation efficiency by over 15% [2]. The company stated that its core mission is to "democratize advanced intelligence by continuously lowering customer costs to ensure artificial general intelligence benefits all humanity" [1]. According to the Artificial Analysis Intelligence Index v4.1, the GPT-5.6 Luna model now offers superior cost-effectiveness compared to DeepSeek's V4 Pro on a per-task completion basis [1]. Additionally, OpenAI projects that the Auto-review model costs within ChatGPT and Codex will decrease to approximately one-tenth of their previous levels [2].
