Two frontier releases in 24 days, and no rulebook yet

On July 24, Anthropic released Claude Opus 5 at $5 per million input tokens and $25 per million output, 24 days after Sonnet 5. The company says the model “more than doubles Opus 4.8’s performance at a lower cost per task” on its hardest agentic benchmark and “comes close to the frontier intelligence of Claude Fable 5 at half the price.” Two frontier-class releases in under a month no longer count as an event. That is now the operating tempo of the industry your vendors build on.

Set that tempo against the regulatory calendar. The revised model-risk guidance took fifteen years to replace its predecessor, explicitly excludes generative and agentic AI, and the RFI meant to start closing that gap has not yet been issued. In the time since the guidance published in April, the frontier has shipped at least two major models. The clock speeds are not close, and they are not converging.

Tempo as a risk parameter

Your model-governance cycle is annual. Your vendors’ model supply chain now turns over monthly. The risk is that ratio, more than any single release.

A monthly cadence plus falling prices creates constant swap pressure. Every few weeks, each vendor with an AI feature gets a fresh reason to change the model underneath you. The reason may be cost or capability, or a version may have been deprecated. A governance process that assumes model stability between annual reviews is now assuming something false.