Anthropic risk report reveals Model 2, unreleased AI model with superior performance over Claude Mythos 5

Anthropic disclosed in its August 2026 risk report the existence of an artificial intelligence model called Model 2, which demonstrates superior performance to Claude Mythos 5 across multiple tasks. Although not yet released publicly, internal testing using the CoBench v2 benchmark demonstrates the performance improvements. The capability increments are modest and focus mainly on overall model enhancement. The report also mentions the existence of a Model 1, leading users to speculate that Model 2 would be its direct successor, possibly as part of further iterations of Claude Mythos Preview.