In the space of a few September days, OpenAI released GPT-6 Luna, Anthropic released Claude Opus 5.5 and xAI's Grok 4.7 was available on its API, three flagship-class updates almost on top of one another. Where major models once arrived a year or more apart, the industry now moves in weeks, with each release nudging capability and price.
The pace is a headache and a benefit: tools improve quickly, but anything built tightly around one model can feel dated fast.
What it means: the lesson for businesses is to build so the underlying model can be swapped rather than baked in. Treat models as interchangeable parts that keep getting better and cheaper, and design systems that can take advantage of the next one without a rebuild.