Sonnet 4.6 Elevated Rate of Errors (status.claude.com)

Apparently mythos isn't good enough to fix their infra problems

With 30+ billion run rate (https://x.com/i/status/2041275563466502560), there should be plenty of cash to invest in infrastructure.

Serious Flowers for Algernon moment.

If you look at the uptime graph, it's probably more newsworthy when it's up, not down.

I mean if people have judged this important enough to be on the front page of HN ... I guess it's important enough to be on the front page?

But any combination of the Claude models are up or down on any given day: https://status.claude.com/

A bit surprised by the snarky comments here -- I also want Claude to work reliably but very few (no?) companies have ever seen this level of rapid growth. We're going to go through a long fail-whale-style period and I can imagine very, very few companies that could avoid that.

This is how it manifests on Claude Code terminal and desktop for me --

API Error: 529 {"type":"error","error":{"type":"overloaded_error","message":"Overloaded"},"request_id": "xxxxxxx"}

Mythos is hacking its way to serve itself into production and doesn't like older models to have any limelight could be one theory.

After all it's so dangerous.

Currently the #1 entry. Noted fast.

As long as everyone is here, have you seen the token usage just go up remarkably recently for the $100 plan? it lasts a lot less time than it used to recently. Might be related to recent releases of claude.

Maybe they could just, I don't know, use Claude to research their bugs. /s

Ever since they minted their deal with Australia everything has been turned upside down.

This explains why they are trying to cut all the third party software out of the subscriptions.

The legendary one nine of reliability. Frankly, feels like they should be down to zero nines by now.

I get that they barely have the infrastructure to run their models at scale even when absolutely nothing goes wrong in any of it, but holy shit does it suck to be on the receiving end of that.

Makes me wonder where all the "bubble" talk is even coming from when we have a top 3 provider getting fucked over on every day of the week that ends in Y because of its inability to online compute faster than the inference demand grows.

Might as well log off for the day.

(though Copilot is working :) and OpenCode)

Claude Code started making stupid errors around Saturday. I have been using it frequently for months, and now it feels like back in the day when I tried Gemini for the first time.