OpenAI cut GPT-5.6 Luna by 80 percent and Terra by 20 percent, effective July 30. Luna now costs $0.20 per million input tokens and $1.20 per million output tokens, down from $1.00 and $6.00. Terra fell to $2.00 and $12.00. Sol did not move. [1]
The paper’s Sunday brief on the Luna cut registered the number. Monday’s fact is the split: the cheap model got cheaper; the flagship grew a Fast mode at twice Sol’s price for up to 2.5 times the speed. [1]
OpenAI says Luna now matches models that were frontier-class a year ago at roughly six cents on the dollar per task, at nearly nine times the speed. It credits production-kernel work that cut serving cost about 20 percent and speculative-decoding gains above 15 percent. [1] VentureBeat placed the move in an explicit price war with Chinese and Google flash tiers. [2]
Cognition posted that GPT-5.6 now “sits on the pareto curve of price/performance efficiency.” That is the X frame: a chart sliding left, not a press release about abundance. [2]
Sol Fast replaces Priority Processing. Requests tagged priority roll into it. Subscription list prices for ChatGPT and Codex stay put; Luna and Terra simply burn fewer credits. AWS pricing was still rolling out when the company posted. [1]
The honest caveat is in the table OpenAI did not discount. Hard reasoning still costs Sol money. Volume work is what got the 80 percent. Accessibility and quality are not the same product. Monday’s market is buying the first and pretending it bought the second.
Replit’s Michele Catasta called Luna “the closest we’ve come to intelligence too cheap to meter.” Notion said Terra matched GPT-5.5 quality at half the cost and 60 percent less time on scoped tasks. Ramp made Luna its default for background agents. Those quotes live on OpenAI’s page. They are advertisements that happen to be specific. [1]
Three weeks after launch is a short life before a reprice. OpenAI did not call it a promotion. It called it kernels and decoding. Competitors still undercut Luna on raw tokens. What changed is that an OpenAI frontier-series name now sits in the bargain bin. That is the war. Sol Fast, at a premium, is the refusal to fight it at the top. [1][2] Trade notes repeated the same table on July 31: Luna 80 percent, Terra 20, Sol unchanged, Fast at 2x. When every desk reprints the spreadsheet, the argument has already moved from whether the cut is real to what it is for. [3] Developers will reroute overnight. Finance teams will take a quarter. The 80 percent is real. The quality story will be argued in evals, not in the price list. Cheap tokens do not make a careful model. They make a frequent one.
-- THEO KAPLAN, San Francisco