Theo: Luna after the 80% price cut is "basically free"

What's this about?

Theo (t3.gg) says the Luna model, after an 80% cost reduction, is cheap enough to use lavishly — he's rebuilding T3 Code's title generation around it and wants to run it on *every* prompt to auto-generate descriptions, feedback, and status updates.

What exactly changes when a model gets "too cheap to meter"?

Below roughly a cent per call, you stop rationing. Instead of one careful model call, you sprinkle dozens of little ones through a product — labels, summaries, suggestions — because each is effectively free. It shifts model use from "a feature" to ambient plumbing, and it's the same economic flip swyx invokes about ontologies today.

What's the community saying?

1. The design-pattern point resonates: near-free inference changes *where* you put AI, not just whether you use it.
2. Skeptics note "Luna" is loosely specified — provider/pricing aren't confirmed in the tweet, so treat the specifics loosely.
3. Cost-per-token as the headline spec again — the same theme as Cursor's efficiency numbers and DeepSeek reseller markups this week.