Anthropic's Internal Model 2 Outperforms Claude Mythos 5

Anthropic is running an internal AI model called Model 2, which performs better than its publicly released top model, Claude Mythos 5. However, the company has no plans to launch Model 2 externally. Gigazine reported on August 17th that Anthropic disclosed these details in an AI risk report. The report describes Model 2 as having "slightly higher capabilities" than Claude Mythos 5 and being widely used internally. For public service, Anthropic released Claude Fable5, which includes additional safety measures. Anthropic employees found Model 2 "clearly improved over Claude Mythos 5 in many tasks," but the improvement was limited, not a large leap. In internal benchmarks, Model 2 was narrowly ahead. However, in some safety evaluations, like a secret mission test, Model 2's success rate was lower than Claude Mythos Preview with extended reasoning. This disclosure reveals Anthropic's product structure, showing a separate line of high-performance internal models. The company prioritizes safety control over raw performance for public releases. This highlights the ongoing tension between AI capability and safety in development.