OpenAI aggressively cuts prices on GPT-5.6 models

OpenAI has announced significant price reductions across its latest GPT-5.6 lineup. The Terra model received a 20% price cut, while the Luna variant achieved a far more dramatic 80% reduction, fundamentally reshaping the competitive landscape for cost-efficient AI models. The company leveraged GPT-5.6 Sol to drive these efficiency gains. The Sol model was deployed to optimize load balancing and, more notably, to enhance the inference optimization process itself. By refining the model's forward pass—the computational stage that converts inputs into token predictions—the system identified inefficiencies that kept GPUs underutilized, including excessive memory movement and suboptimal data layouts. Particularly noteworthy was GPT-5.6 Sol's autonomous rewriting of production kernels, the core code executing the model's mathematical operations. This capability was enabled by prior training of GPT-5.6 on kernel development and optimization using Triton and Gluon, two open-source GPU programming frameworks maintained by OpenAI. The combined optimization efforts achieved a 20% reduction in end-to-end serving costs. Luna's new pricing of $0.20 per million input tokens and $1.20 per million output tokens now undercuts Google's Gemini 3.1 Flash-Lite. The model is also dramatically cheaper than Anthropic's Claude Haiku 4.5, priced at just one-fifth the previous rate for input tokens.