Opus 4.8, Sonnet 4.6, and other Anthropic models are suffering from overload and elevated error rates. Anthropic is working the issue. Occasionally the LLMs will come back online, but there have been multiple drops and interruptions.
This does NOT appear to be an outside attack on Anthropic (via DDS or something). Instead, it seems to be contained to Anthropic servers and related third-party suppliers.
Base Claude.AI, Claude Console, Claude API, and Claude Cowork/Code are all affected. Expect the instability to continue for a while.