OpenAI Cut GPT-5.6's Cheapest Tier by 80 Percent
Luna dropped from $1 to $0.20 per million input tokens. Terra dropped 20 percent. OpenAI calls it advancing the price-performance frontier.
OpenAI just cut its cheapest GPT-5.6 tier by 80 percent. Here's what's actually behind it.
OpenAI cut pricing across its GPT-5.6 lineup on July 30, 2026. Luna, the cheapest tier, dropped 80 percent. Terra, the mid tier, dropped 20 percent. Sol, the flagship, gained a new paid Fast mode instead of a price cut.
The actual numbers
According to reporting from The Decoder and other outlets tracking the change, Luna went from $1.00 and $6.00 per million input and output tokens to $0.20 and $1.20, an 80 percent cut. Terra went from $2.50 and $15.00 to $2.00 and $12.00, a 20 percent cut. Sol's base pricing held steady, but it picked up a Fast mode priced at double the rate for roughly 2.5 times the speed, replacing what OpenAI had called Priority Processing.
What's paying for it
OpenAI frames the change as advancing the price-performance frontier, attributed in coverage to infrastructure efficiency work: roughly a 20 percent cut in serving cost and a 15 percent gain in token-generation efficiency from speculative decoding, some of it produced by Sol itself rewriting parts of the production inference code.
Why outlets are calling this a China pricing move
The Decoder and other coverage frame the Luna cut as OpenAI matching the aggressive per-token pricing that Chinese open-weight labs like DeepSeek and Zhipu's GLM have been running, rather than a claim OpenAI itself made directly in the material available. Worth keeping that distinction straight: the numbers are confirmed, the competitive framing is the press's read on them.
Why a build studio cares
For an AI workflow, the cheapest usable tier is often what decides whether a background agent step runs on the biggest model everywhere or gets routed down to something smaller. An 80 percent cut on Luna changes that math specifically for high-volume, low-stakes agent tasks, worth rerunning any cost model built before July 30.
Next step: read The Decoder's pricing breakdown. If you want a second opinion on model routing costs for an agent workflow, write to us at hello@gattyworks.com.