OpenAI released GPT-6 in two variants on September 22, named Luna and Sol, continuing an industry pattern of shipping a fast, cheaper model alongside a slower, more capable one. Splitting a model line this way lets customers pick speed and low cost for routine work and reserve the heavier model for hard problems.
The two-tier approach has become standard across major labs because most requests are simple and a minority are hard, and paying flagship prices for every request wastes money.
What it means: when a tool offers more than one model, the cheaper, faster option usually handles ordinary tasks such as drafting replies or summarising notes perfectly well. Reserving the expensive model for genuinely difficult work is the simplest way to control an AI bill without losing quality where it matters.