OpenAI slashes GPT-5.6 prices after the model speeds up its own backend

OpenAI just dropped prices on its GPT-5.6 models by up to 80%—partly because the AI helped rewrite its own infrastructure.

OpenAI announced price cuts across its GPT-5.6 family, lowering the cost of its lightweight Luna model by 80% and its balanced Terra model by 20%. In the API, Luna now costs $0.20 per million input tokens, while Terra drops to $2.00. OpenAI also introduced a Fast mode for its top-tier Sol model, which delivers up to 2.5× faster speeds at twice the standard price.

Why it matters: High-volume AI tasks just became far cheaper to run. Luna delivers performance comparable to last year's frontier models at roughly six cents on the dollar and nearly nine times the speed. That makes running multi-step workflows, background code changes, and heavy document processing much easier on developer budgets.

Know this: The price drop came from a tight feedback loop. OpenAI used GPT-5.6 Sol to autonomously rewrite and optimize its own production kernels, cutting model serving costs by 20% and boosting generation efficiency by over 15%.

When the model starts tuning its own backend code, everyone's API bill gets a little lighter.