Anthropic reveals Model 2, which beats Claude Mythos 5 on many tasks

In its August 2026 risk report released on August 15, Anthropic mentioned an unreleased AI model called Model 2. It performs better than Claude Mythos 5, the strongest public model, on many tasks, though the improvement is modest and focused on overall capability. The report includes results from internal benchmark CoBench v2, showing Model 2's advantages. It also mentions Model 1, leading users to guess that Model 2 could be the successor to Model 1, and both might be further iterations of Claude Mythos Preview.