Anthropic flags elevated errors across Claude Mythos, Fable and Opus models

What happened

Anthropic is working to fix an incident causing elevated error rates across a wide slice of its Claude model lineup. The company's status page began tracking the problem on September 3, 2026, and the incident report reached the Hacker News front page shortly afterwards, signalling significant developer interest.

According to the status page, the disruption spans four products: the claude.ai consumer service, the Claude API at api.anthropic.com, the Claude Code developer tool, and Claude Cowork. The impact is therefore not limited to the chat interface; applications and automations built directly on the API are exposed as well.

Which models are affected

The initial report named three models: Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5. Roughly 24 minutes later, Anthropic published what it described as an exhaustive list, which widened the scope considerably: Mythos/Fable 5.1, Mythos/Fable 5, Opus 5, Opus 4.8, and Opus 4.6.

Two things stand out. First, the incident covers multiple model families at once rather than a single deployment. Second, it reaches back across generations, from the 5.x releases down to Opus 4.6. If the list is as complete as the status update claims, any model not on it should currently be unaffected, which gives teams with multi-model routing a potential fallback path.

Timeline of the incident

• 13:26 UTC — Anthropic posts that it is investigating elevated errors on requests to Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5.
• 13:41 UTC — The company reports it has identified the cause, says a fix is in progress, and promises a further update.
• 13:50 UTC — The exhaustive list of affected models is published, adding Mythos/Fable 5, Opus 4.8 and Opus 4.6 to the original three.

Notably, the latest entry on the page is the expanded model list. There is no monitoring or resolved post as of the report's appearance on Hacker News at roughly 14:07 UTC, so the incident should be treated as ongoing, with further updates expected on the status page.

Why it matters

For anyone running production workloads against the Claude API, this is the kind of incident that surfaces immediately as failed requests, timeouts or retry storms. Because Claude Code and Claude Cowork are also affected, the blast radius extends beyond direct API calls into agent-driven workflows and developer tooling that many teams now treat as core infrastructure.

The shape of the incident is worth attention too. Errors spanning two model families and several versions point to a shared underlying cause rather than a bad single-model rollout, although Anthropic has not yet published details of what that cause is. Until it does, the practical options are limited: watch the status page, lean on unaffected models if your routing allows it, and make sure client-side retry and backoff logic is doing its job rather than amplifying load on an already degraded service.

The episode is also a reminder of how concentrated dependencies on a single AI provider have become. One incident page now simultaneously touches a consumer chat product, a widely used API, a coding agent and a coworking tool. Teams building on these platforms need monitoring and fallback plans that assume this kind of concentration, not just per-model dashboards that would show a narrower failure as business as usual.