OpenAI released GPT-6 Astra on 3 September 2026 and followed it with GPT-6 Luna on 22 September, marking the arrival of its next flagship model generation. The two releases within a single month underline how quickly the leading labs are now iterating, with each version pushing capability and adjusting price.
GPT-6 is positioned for the harder end of the market: long, multi-step tasks and agent-style work where a model must plan, act and check its own output rather than answer a single question.
What it means: for a business, the practical lesson is not to chase each release but to build so the underlying model can be swapped. Treat models as interchangeable parts that keep getting better and cheaper, and design systems that can adopt the next one without a rebuild, because another version is rarely far behind.