The acceleration of AI model releases and the rise of recursive self-improvement

On September 23, 2026, two major AI models were announced on Tuesday, with Anthropic and OpenAI releasing updated versions of their flagship models. Anthropic introduced Claude Opus 5.5, which matches the performance of its previous flagship model, Claude Fable 5.1, but at 40% lower cost. OpenAI launched GPT-6 Sol and GPT-6 Luna, which are faster and more affordable versions of its GPT-6 Astra model. SpaceXAI also released Grok 4.7, a new flagship model with enhanced capabilities for coding and knowledge work. The release frequency for frontier models has increased for both Anthropic and OpenAI, with Anthropic releasing eight flagship models this year and OpenAI six. However, the time between genuinely new flagship models has remained consistent. Anthropic reports that its Claude models now lead 26% of its AI R&D efforts, highlighting the use of AI in model development. OpenAI noted that its AI agents are now working 3.1 days for every 1 day of human labor, showcasing the growing role of AI in its operations. Researchers are increasingly using AI to design computing infrastructure, generate synthetic training data, and optimize software frameworks. This trend, known as recursive self-improvement, is reshaping the pace of AI innovation, making product launches a less reliable indicator of technological progress.