OpenAI Cuts GPT-5.6 Prices via Autonomous Model Optimization
OpenAI has announced significant price reductions for its GPT-5.6 Terra and Luna models, cutting prices by 20% and 80% respectively. These savings were achieved by using their own GPT-5.6 Sol model to autonomously optimize inference computation and GPU utilization. The massive 80% price drop makes GPT-5.6 Luna highly competitive, undercutting rivals like Anthropic's Claude Haiku 4.5 and Google's Gemini 3.1 Flash-Lite on key pricing metrics. This demonstrates a growing trend of using AI models to self-optimize their own underlying code and serving infrastructure. GPT-5.6 Sol optimized the forward pass by rewriting production GPU kernels in Triton and Gluon, reducing end-to-end serving costs by 20%. Luna's new pricing stands at $0.20 per million input tokens and $1.20 per million output tokens.
## BACKGROUND
In deep learning, the forward pass is the computation phase where input data travels through the neural network's layers to generate a prediction or output token. Inference optimization focuses on improving the speed and resource efficiency of running these models, often by minimizing GPU idle time caused by memory movement and synchronization bottlenecks.