Anthropic has been using an internal AI model named 'Model 2' for various tasks, including coding and research. According to the company's August 2026 Risk Report, Model 2 outperforms every publicly available version of Claude. The report places the model in the Mythos class and states it is slightly stronger overall than Claude Mythos 5, though weaker in some areas. It does not show a significant capability jump compared to the transition from Opus 4.6 to Mythos. On the company's internal capability index, AECI, Model 2 scores about 1.5 points above Mythos 5, which is a smaller gain than the jump from Mythos Preview to Mythos 5.
Internally, Anthropic relies heavily on Model 2 for coding, data generation, and research and engineering, often through continuously running agents. While Claude now writes most of the code in production systems, Model 2 underwent internal review before deployment but was not tested as thoroughly as Mythos 5. The company found no new or more concerning misalignments in the process and rates the overall risk from misalignment as 'low.' There are currently no plans to release the model externally.
Anthropic's ECI is a collection of internal benchmarks. 'Model 2' is said to land 1.5 points above Mythos 5, but it isn't plotted here. | Image: Anthropic
Source: thedecoder
