OpenAI has cut the price of GPT-5.6 Luna, the entry model of the GPT-5.6 family, to $0.20 per million input tokens and $1.20 per million output tokens. Luna launched at $1 per million input tokens — the new figure is an 80% reduction on the input side.
What changed
The three-model lineup now prices out as follows: Luna at $0.20 / $1.20, Terra at $2 / $12 and Sol at $5 / $30 per million tokens. Only Luna moved; Terra and Sol are unchanged.
Batch processing runs at half the standard rate across all three, and long-context requests are billed at roughly double the short-context rate — so the effective price depends heavily on how a workload is shaped, not just which model is picked.
Where this sits in the market
Luna at $0.20 input lands in the same bracket as DeepSeek V4-Flash at $0.14 and Google's Flash-Lite tier. Twelve months ago the cheapest credible frontier-adjacent option cost several dollars per million input tokens. The floor has moved by an order of magnitude, and it moved fastest in the last quarter.
Why it matters
For anything high-volume — classification, extraction, routing, first-pass summarisation — the model bill stops being the constraint at these levels. That changes which products are worth building: workflows that were uneconomic at $1 per million inputs become viable at $0.20.
The flip side is that price alone no longer separates the cheap tier. When four vendors sit within a few cents of each other, the decision moves to latency, context handling, tool-calling reliability and how the provider behaves when it needs to raise prices later.
What to watch next
A cut this size usually signals capacity rather than generosity. Inference gets cheaper when a provider has spare hardware and an efficient serving stack, and prices follow. The question for anyone budgeting on these numbers is whether the floor holds — DeepSeek has already said publicly that its own prices are going back up.
The practical move is to treat the cheap tier as interchangeable. Keep prompts portable, avoid leaning on provider-specific features for the highest-volume paths, and re-benchmark quarterly. At these prices the switching decision should be driven by quality on your own evaluation set, not by a table of list prices that changes every few weeks.