Anthropic investigates a spike in errors across five Claude models

What Anthropic posted

On October 25, 2026, at 06:14 UTC, Anthropic published an incident on its Claude Status page, the public dashboard that shows whether its services are working. The incident is titled "Elevated errors for multiple models." "Elevated errors" simply means more requests are failing than normal.

As Unite.AI reported, the post says Anthropic is looking into failures on requests to five models. They come from the Mythos, Fable, Opus, Sonnet and Haiku lines, with version numbers 5.1 or 5.5. Anthropic promised another update as soon as it can.

When the report was written, the incident was still marked Investigating. That is the earliest stage. It means the company has acknowledged a problem but has not yet said what caused it or how it plans to fix it.

Which services are affected

The record lists four affected services: the claude.ai website and app, the Claude API, Claude Code and Claude Cowork. The API is the connection other software uses to talk to Claude, so apps built on top of it can fail too, even if you never open Claude directly.

All four were marked Partial Outage. That label suggests some requests are failing, not that everything is down. Two other components, Claude Console and Claude for Government, were still shown as Operational.

The page also shows uptime over the past 90 days. For claude.ai it reads 99.58 percent. Claude for Government shows 100 percent. Those figures describe the recent past, not today's problem, but they give a sense of how often each service has stumbled lately.

Not the first wobble this week

The status page history shows a run of short incidents just before this one. Three were resolved on October 24, 2026. One covered Fable models, and Anthropic's record says it identified the cause and then deployed a fix with monitoring in place. Another hit Opus 5.5, and a third touched Fable 5.1, Sonnet 5.5 and Haiku 5.5.

One update on that third incident also mentioned fast mode requests to Opus 5.5. A later update said success rates were returning to normal for all the affected models.

Earlier entries show a similar pattern. On October 22, an incident hit Connectors, which affected claude.ai and Claude Cowork. On October 21, a separate incident described degraded performance for Claude Fable 5. Another, on October 13, involved usage data not loading on platform.claude.com, and Anthropic said API requests were not affected.

The source also points to an August 20, 2026 incident with a similar title, which touched the same four services. Nothing in the source says these events are linked. It is simply a record of how often small disruptions appear on the page.

What this means in practice

Because five models are named, switching from one Claude model to another may not help. The problem looks wider than a single model. That is an inference from the incident wording, not something Anthropic has confirmed.

Practical steps are simple:

• Check the status page before you change your own settings or blame your account.
• Keep drafts and prompts saved outside the chat window so a failed request never costs you work.
• If you run a tool built on the API and it suddenly fails, look at the provider's status first.
• Have a manual fallback for urgent tasks, such as saved templates or a second assistant.

Anthropic has not published a cause, a scope of affected users or an estimate of when things will be fixed. Until it does, treat any guess about the reason as just that.