Anyone trying to work with Claude this morning ran straight into elevated error rates. Anthropic has been reporting disruptions since early today, and they don’t hit just one corner: claude.ai, the API, Claude Code and Cowork are all in it, across models from Opus 4.8 through Opus 5 to Fable 5 and Mythos 5.
The tone on Hacker News matched. “Anthropic has constant failure since morning,” one user wrote flatly. Others asked how a service goes down globally all at once – do all regions really run on the same infrastructure?
Not a one-off
The annoying part isn’t the single outage. Outages happen; every big service has them. The annoying part is the frequency. In August alone there were noticeable disruptions on the 5th, 12th, 13th, 16th, 18th and 20th – and now the 24th. That’s not a bad streak anymore, that’s a pattern.
Nobody outside knows the exact cause. The obvious guess: demand is growing faster than the infrastructure can keep up. Auto mode, subagents, long-running coding sessions – all the things Anthropic itself encourages – pull serious load. But the best recommendation doesn’t help much when the lights keep going out.
My take
I like Claude, and I’m sticking with it. But reliability isn’t a nice extra – it’s the foundation you build a tool into your workday on. Put a pipeline on a model that wobbles every couple of weeks, and you get nervous. Rightly so.
For you, the practical takeaway is simple: have a plan B. A second model, a local fallback, or just the calm to go do something else for a bit. And for Anthropic, hopefully the takeaway is: get the base stable first, before the next big announcement lands.
Sources: