OpenAI Cuts GPT-5.6 Prices. Models Become Commodities

A Substantial Discount Arrives Unusually Fast

OpenAI is dropping the operation costs of two of its newest models across the board. The price for 1 million tokens on the lightweight GPT-5.6 Luna drops by 80% (to 1 dollar for input and 6 dollars for output). The change comes less than a month after the family's launch.

Price per Token Shapes Agents

Cheaper inference changes how models are integrated into corporate systems. If an agent must autonomously scan long documents and repeatedly call tools, token consumption grows. Luna's 80% discount means room for longer draft thinking without exhausting the budget.

Competition from Below Forces Concessions

This move shows the pressure from open-weight models and fierce competition offering comparable performance at a fraction of the price. If smaller models can handle 90% of a standard workload, the more expensive ones must justify their existence on specialized tasks.

Accounting Decides Deployment

The most important future signal will be the rate of model adoption in automated processes. If the cheaper Luna reliably replaces more expensive variants in standard routing, inference pricing will definitively surpass absolute benchmark positions.

Lilith's verdict

Price has long represented a product capability, not merely an invoice item. OpenAI is cutting prices out of fear that customers won't use the most expensive model as a gatekeeper for every trivial query.