AI model update cycles shrink from 125 days to 44 days
An analysis says the gap between high-performance model updates at major US and Chinese labs has fallen to about 44 days, naming releases such as GPT-6 Astra in September.
AI USED TO UPDATE IN MONTHS. NOW IT’S MOVING IN WEEKS. Something interesting is happening across the AI industry. A new analysis of major U.S. and Chinese AI labs found that the average gap between high-performance model updates has fallen from 125 days between 2023 and March 2026, to just 44 days from April to September 2026. That’s roughly one major update every month and a half. And you can actually see it happening. ✧ OpenAI released GPT-6 Astra in September. ✧ Anthropic released Claude Fable 5.1 and Mythos 5.1. ✧ Google pushed Gemini 3.8. ✧ Meta pushed Muse Spark 1.3. ✧ DeepSeek has also been releasing updates at a much faster pace. But here’s the part that really caught my attention: AI is increasingly helping build AI. Anthropic says that as of August, AI led 26% of its measured AI R&D work, while more than 90% involved AI collaboration. OpenAI reported 3.1 AI agent-workdays for every human researcher workday in August. So the cycle starts looking like: Better AI → faster research → faster experiments → better AI → even faster research. We’re obviously not at the Ultron building an army stage 😂 But the basic idea of AI becoming part of the machinery used to build the next generation of AI? That’s already happening. And if model development keeps compressing from months to weeks the real question isn’t just how powerful AI gets. It’s how quickly the next version can arrive. That might be the most important AI race nobody is talking about enough.