OpenAI 于 7 月 30 日宣布调整模型定价，其中 GPT-5.6Luna 模型降价 80%，Terra 模型降价 20%[1]。这一举措反映了该公司商业战略的重大转变，从强调旗舰模型性能优势转向推广中低端产品。
根据 OpenAI 的建议，用户应根据任务复杂程度分层使用不同模型[1]。具体来说，复杂任务应先由 GPT-5.6Sol 完成需求分析和方案设计，随后交由 Luna 执行、编写代码并运行测试[1]。OpenAI 建议开发者根据"任务的重要程度、出错代价、紧急程度和规模"来匹配相应模型[1]。
在技术优化方面，Sol 模型通过参与底层代码优化并进行模型自主重写生产内核，使端到端运行成本下降 20%，Token 生成效率提升 15%[1]。这一定价调整延续了过去两年半的降价趋势：GPT-4 首发时为 30 美元/百万 Token，GPT-4o 后降至 5 美元，GPT-4o mini 进一步降至 0.15 美元，Luna 模型价格现已接近 mini 廉价区间[1]。
OpenAI announced significant price reductions on July 30, restructuring its model pricing strategy to emphasize cost-effective alternatives over flagship offerings [1]. The GPT-5.6 Luna model saw an 80 percent price cut, while the Terra model dropped 20 percent [1]. This pricing adjustment reflects a strategic pivot toward promoting mid-to-lower-tier models rather than relying on premium model sales [1].
The company now recommends a tiered approach where developers match models to task requirements [1]. For complex assignments, users should employ the GPT-5.6 Sol flagship model to conduct requirement analysis and design planning, then delegate execution tasks—including code writing and testing—to the cheaper Luna model [1]. OpenAI suggests developers consider four factors when selecting models: task importance, cost of errors, urgency, and scale [1].
Technical optimizations accompany the pricing shift, with Sol model improvements achieving a 20 percent reduction in end-to-end operational costs and a 15 percent increase in token generation efficiency through autonomous kernel rewriting [1]. Over the past two and a half years, OpenAI's pricing trajectory has been dramatic: GPT-4 launched at 30 dollars per million tokens, GPT-4o descended to 5 dollars, and GPT-4o mini reached 0.15 dollars, with Luna now positioned in the low-cost tier alongside mini [1].
