OpenAI has reduced the cost of GPT-5.6 Luna by 80%, bringing the input price to $0.20 per million tokens and the output price to $1.20. This specific price adjustment places Luna below Google’s Gemini 3.1 Flash-Lite for both input and output tasks. The company attributes this shift to GPT-5.6 Sol, an internal model that optimises inference by rewriting production kernels in Triton and Gluon. This autonomous process identified excess memory movement and inefficient data layouts within the forward pass. By precomputing work and parallelising operations, the team reduced end-to-end serving costs by 20%.
The reduction alters the market positioning for lower-priced models relative to competitors like Anthropic and Google. Luna is now significantly cheaper than Claude Haiku 4.5 for input operations. Developers can switch existing applications to this model without incurring higher costs.
* Input cost is $0.20 per million tokens
* Output cost is $1.20 per million tokens
* Serving costs fell by 20% via kernel optimisation
