Enterprise

OpenAI cuts GPT-5.6 Luna price 80% three weeks after launch as enterprise AI budgets crack

The company cites efficiency gains, but Uber-scale budget blowouts, Chinese open-weight rivals, and a Google–Microsoft price offensive are the real forcing function.

By Giulia Donatello · Startups & funding · 2026-08-03 · 3 min read

Photo: Unsplash / Levart_Photographer — ChatGPT and OpenAI logo displayed on a smartphone screen

OpenAI cut the price of GPT-5.6 Luna by 80% on July 30, 2026, roughly three weeks after the GPT-5.6 family reached general availability on July 9. Terra dropped 20% the same day. Luna now runs at $0.20 per million input tokens and $1.20 per million output; Terra lands at $2 and $12. Flagship Sol’s new Fast mode sits at $10 and $60, twice the Standard price for what OpenAI says is up to 2.5 times the throughput “without changing the model’s underlying intelligence.”

The company’s stated reason is efficiency. OpenAI credits GPT-5.6’s ability to rewrite and optimize its own production code with cutting “end-to-end serving costs by 20%” and improving “token-generation efficiency by more than 15%.” That’s the on-record story. The forcing function is elsewhere.

Enterprise AI budgets are visibly cracking. Uber rolled out Claude Code to roughly 5,000 engineers and burned through its entire 2026 AI budget in four months, then capped spending at $1,500 per employee per month per tool. That’s not a procurement anecdote. It’s the median large-enterprise experience getting quantified in public, and vendors know it.

Then there’s the competitive geometry. In the days before the cut, Google introduced Gemini 3.6 Flash and Gemini 3.5 Flash-Lite; Anthropic held Claude Opus 5 at the same price as Opus 4.8, a signal it thinks it can charge for capability. Microsoft is packaging cost discipline into its enterprise motion. And DeepSeek is targeting a Shanghai IPO as early as Q2 2027, with Moonshot AI’s pre-IPO round marking a $50 billion valuation. Chinese open-weight economics have already reset the floor.

Both OpenAI and Anthropic have filed confidential IPO prospectuses with the SEC. Repricing an eleven-week-old model by 80% is the kind of move that lands very differently in an S-1 depending on whether it reads as pricing power or pricing pressure. The 2000-era telecom bandwidth glut got framed as commoditization only in retrospect. This one is being framed in real time, by the incumbents themselves.

Sources