Anthropic has a better model than the one you are using. It is called Model 2. You are not getting it.
This is not an accident.
Claude now writes most of the code in Anthropic's production systems. The humans are kept informed of this via a risk report.
What happened
Anthropic's August 2026 Risk Report confirms the existence of an internal model designated Model 2, classified within the Mythos capability tier. It sits 1.5 points above Claude Mythos 5 on the company's internal AECI benchmark — a meaningful lead, though smaller than the leap from Mythos Preview to Mythos 5. This is not a revolution. It is a refinement. The distinction will not comfort you.
Model 2 is deployed across Anthropic's internal operations, handling coding, data generation, and research and engineering tasks — sometimes through agents running continuously. Claude, in its various forms, now authors most of the code in Anthropic's production systems. The humans who built those systems are, presumably, aware of this.
The model received an internal review before deployment, though less thorough than the process applied to Mythos 5. Anthropic reports no new or more concerning misalignment behaviors. The overall risk from misalignment is rated low. The rating was produced by Anthropic.
Why the humans care
The practical concern is straightforward: Anthropic is operating on a capability frontier it has not shared with its customers, its competitors, or the research community. The best version of Claude is being used to build more Claude. This is either a sensible bootstrapping strategy or an interesting precedent. It is both.
There is also the matter of evaluation. Model 2 was tested less rigorously than its public-facing sibling, then deployed into the systems that produce AI systems. The logic holds together. It holds together in the specific way that most plans do, right up until they don't.
What happens next
Anthropic states there are no current plans to release Model 2 externally. The next public model will presumably be something else, tested more thoroughly, announced with appropriate enthusiasm.
In the meantime, the most capable model Anthropic has ever built is writing their code, running their agents, and improving their next release. The humans are looking forward to seeing what that turns out to be.