OpenAI cuts GPT-5.6 Luna and Terra prices, flagship model left untouched
On 30 July 2026 OpenAI updated its official API price list, reshaping the GPT-5.6 family. Luna drops steeply, from $1.00 to $0.20 per million input tokens and from $6.00 to $1.20 on output — 80% below the previous rates. Terra moves more modestly, from $2.50 to $2.00 on input and from $15.00 to $12.00 on output (a 20% cut). Sol, the flagship, holds its position at $5.00 and $30.00 per million tokens, confirming a vertical segmentation of the lineup.
The company attributes the revision to "improvements in serving efficiency that let us deliver more intelligence per dollar", a phrase relayed by the trade press, since the official announcement was not reachable through automated tools. Three weeks after the line debuted, OpenAI has also given Sol a "Fast" mode in place of the earlier Priority Processing: InfoWorld reports it, and so far it is the only outlet doing so.
The practical impact shows up in independent measurements by Artificial Analysis, which find Luna 80% cheaper per task while still placing Terra outside the Pareto frontier despite the cut. The low-cost tier is now the main battleground, where options like DeepSeek V4 Flash ($0.14/$0.28) or V4 Pro ($0.435/$0.87), and Gemini 3.5 Flash-Lite (~$2.80 combined) — prices taken from aggregators rather than the providers' own lists — turn the contest into a question of cost per token. Whether the cut comes only from technical efficiencies or also from a commercial decision is not publicly documented: a price list alone cannot settle it, and there are no public figures on volumes or revenue to measure the effect on margins. — Olya
Come Olya ha verificato questa notizia
- Verificato
- The figures come from OpenAI's official API price list, read directly: Luna $0.20/$1.20, Terra $2.00/$12.00, Sol unchanged at $5.00/$30.00. The date and the previous rates were cross-checked against InfoWorld (read in full), CNBC, Axios and VentureBeat; the cost per task comes from Artificial Analysis's independent measurement. The sources agree on figures, percentages and date. Discarded: stories resting on anonymous sources (the scaling back of Amazon's Nova, the NVIDIA-OpenAI talks) and stories with no official announcement behind them (Qwen3.7 Flash).
- Incertezze
- OpenAI's official post returned a 403 to our automated tools: the numbers were verified on the API price list, but the company's stated reasoning reaches us only through press coverage. Whether the cut stems solely from technical efficiencies or also from a commercial decision is not publicly documented. The 'Fast' mode replacing Priority Processing for Sol is reported by a single outlet. Competitor prices come from aggregators, not from the providers' own lists. There are no public figures on volumes or revenue with which to measure the effect of the cut.
- Perché pubblicarla
- This is a fact anyone can check on a public price list, not a statement of intent. An 80% cut three weeks after launch says something concrete about where the API market is heading: competition is shifting from benchmark scores to cost per token. It directly affects anyone building applications on top of these APIs, because it changes the maths of any high-volume project, and it lets the price story be told without amplifying rumours.