Anthropic's "2026 August Risk Report" released on August 15 accidentally revealed an upcoming AI model called Model 2. The report states that this model outperforms the currently available strongest model, Claude Mythos 5, in multiple tasks.

The improvement is not significant, but it focuses on the refinement of comprehensive capabilities across generations

According to the disclosed details, compared to Claude Mythos 5, Model 2 shows significantly stronger performance in multiple tasks, but the extent of the improvement is relatively limited, mainly reflecting an overall enhancement of AI's comprehensive capabilities. The report also includes comparison results from Anthropic's internal benchmark test "CoBench v2" as evidence.

image.png

The report also mentioned the term "Model 1," leading users to speculate that Model 2 might be a subsequent version of Model 1, and both could be further iterations of Claude Mythos Preview. If true, Anthropic is quietly advancing its capability ladder with internal numbering, while public naming lags behind the actual progress.