The pace of AI model releases has accelerated significantly. Between early 2024 and September 22, OpenAI and Anthropic released 14 flagship models, averaging one every 19 days. Domestic Chinese manufacturers released at least 28 models in the same period, averaging one every 9 days. Combined, a new flagship model appears every 6 days.
Anthropic reports that AI is increasingly involved in its own development. By August, Claude handled 26% of AI research tasks. At OpenAI, AI agents performed 3.1 times the workload of human researchers by mid-August. This automation contributes to the rapid release cycle, with Sonnet 5.5 and Haiku 5.5 expected in the coming weeks.
Frequent updates create challenges for developers, as legacy prompts and parameters often become obsolete. Official guidelines now advise removing detailed process instructions, as newer models like Opus 5.5 perform self-checks automatically. Improper migration of settings can lead to errors or increased operational costs.
Performance benchmarks are being surpassed rapidly. The ARC-AGI-3 test, which saw top models scoring 0.37% in March, was dominated by GPT-6 Astra by September 3, achieving 62.7% and reaching 99.9% with specific frameworks. Newer models like Sol and Opus 5.5 now lead coding and writing leaderboards with record-breaking scores.
