AI Pulse · Business Models & Winners

OpenAI dropped the price of GPT-5.6 Luna by 80% this week, bringing it down to $0.20 per million input tokens and $1.20 per million output tokens. That is a permanent cut from the list price set back in July. At the same time, they killed Priority Processing and replaced it with a “Fast” tier that costs twice as much as standard. One move makes Luna cheaper. The other makes speed more expensive. Both happened in the same pricing cycle.

The timing is not random. Anthropic’s promotional pricing, which has been undercutting OpenAI for months, is about to expire. OpenAI is clearly positioning Luna as the volume workhorse while pushing anyone who needs low latency into a higher-margin tier. The result is a 25x spread between OpenAI’s cheapest and most expensive options, which means routing logic now matters more than picking the right model. If you are running AI at any real scale, you cannot just point everything at one endpoint anymore. You need to decide what goes where, and you need to do it automatically.

This is not abstract. If you are processing customer support tickets, drafting emails, or running document analysis, the difference between routing intelligently and routing lazily is real money every month. The full pricing tracker lays out where every major provider stands right now, and the picture is messy. OpenAI is cheaper on volume, Anthropic is still competitive on promo rates until they are not, and the fast lanes cost more everywhere.

The companies building AI command centres, the kind of system we put together in the Omni Command Centre, are routing tasks across models and tiers based on cost, speed, and accuracy in real time. That is not a nice-to-have anymore. It is table stakes if you are running more than a few thousand calls a month.