7 ms·
Sonnet 4.6 Elevated Rate of Errors
- capnsketch 5mo agoApparently mythos isn't good enough to fix their infra problems
- kubb 5mo agoIt’s just so POWERFUL and DANGEROUS that its very aura disrupts the weaker models.
- tristanj 5mo agoIt's a lack of compute. Anthropic is growing double digit % every month, and they're growing faster than they can acquire compute resources. Plus, they do not want to overbuild computer, like what OpenAI is doing.
- NiekvdMaas 5mo agoWith 30+ billion run rate (https://x.com/i/status/2041275563466502560 https://x.com/i/status/2041275563466502560), there should be plenty of cash to invest in infrastructure.
- arcfour 5mo agoMoney pays for infrastructure. It doesn't will infrastructure that doesn't exist into existence.
- 0123456789ABCDE 5mo agothere is enough money; there isn't enough infrastructure/hardware where you can spend that money.
- rbmck 5mo agoSerious Flowers for Algernon moment.
- ak4153 5mo agoWhich side the getting smart or dumbing down
- taspeotis 5mo agoI 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/ https://status.claude.com/
- jonatron 5mo agoIf you look at the uptime graph, it's probably more newsworthy when it's up, not down.
- tao_oat 5mo agoA 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.
- rvz 5mo agoHow can Claude work reliably if Claude keeps going on vacation for several hours? Maybe it is recovering from the weekend a few days ago, but wanted to take an extra day off like it did on Monday, hence the "outage".
- ben_w 5mo ago> How can Claude work reliably if Claude keeps going on vacation for several hours? Not that I wish to anthropomorphise it in this answer, but businesses have managed just fine when humans do this for "lunch breaks" and "going home for the evening to sleep". (And even mandatory meetings which should have been emails).
- sassymuffinz 5mo agoThing is, if Dave the programmer goes on vacation or calls in sick for the day, hopefully you have a larger team to fall back on and your business doesn't grind to a halt. No one is apparently noticing that if they build their entire business model around AI being a certain price and availability they're essentially building one giant point of failure into their productivity. What if the price shoots up 10x or Claude goes down for a day, or what if he's occasionally drunk (hallucinating). Reliability is sometimes a more important facet of business than ultra speed and productivity.
- ben_w 5mo agoAye, correlated failure is not something to be overlooked. Mistaking correlated risk for uncorrelated risk was a big part of the global financial crisis. There are fallback mechanisms when the risk is per model provider (as in, "What if the price shoots up 10x or Claude goes down for a day" is a manageable concern), but I'd be more worried about the way all models regardless of provider have similar failure modes, i.e. that some tasks fail in similar ways for all models. In some ways, LLMs are collectively like Star Trek's Borg: you've met one, you've met all of them.
- albert_e 5mo agoThis 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"}
- wg0 5mo agoMythos 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.
- tipiirai 5mo agoCurrently the #1 entry. Noted fast.
- ApolloRising 5mo agoAs 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.
- gambiting 5mo agoI'm on the basic £18/month plan and with Sonnet 4.6 I literally get 20 maybe 30 minutes of use out of it per day. It's borderline useless now. I was using it for some Home Assistant changes yesterday and it used up my entire daily allowance after 8 prompts.
- ManlyBread 5mo agoI guess 2026 is the last year AI is widely available to anyone who isn't willing to shell out hundreds if not thousands for a monthly subscription. I guess all that's left is to thank all the investors for the free ride LOL
- andrewinardeer 5mo agoGood luck jacking prices up too high with new open models flying around daily.
- baq 5mo agogood luck getting mythos/spud quality models open
- _delirium 5mo agoHard to square that with how good open-weights models are getting? I'm doing stuff with Qwen3.5-4b that required a frontier hosted model less than a year ago.
- baq 5mo agothe problem is you're still a year behind with this approach and it isn't at all clear locally hosted models can keep the gap. need more turboquant-like algorithmic boosts for this to happen.
- pjmlp 5mo agoMaybe they could just, I don't know, use Claude to research their bugs. /s
- OhioMan2943 5mo agoEver since they minted their deal with Australia everything has been turned upside down.
- crimsonnoodle58 5mo agoI'd say it was when OpenAI had a mass exodus due to them making a deal with the Department of War (which they then backtracked on [1]). This started the QuitGPT movement [2]. [1] https://www.bbc.com/news/articles/c3rz1nd0egro https://www.bbc.com/news/articles/c3rz1nd0egro [2] https://quitgpt.org/ https://quitgpt.org/
- gdorsi 5mo agoThis explains why they are trying to cut all the third party software out of the subscriptions.
- ACCount37 5mo agoThe 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.
- anshumankmr 5mo agoMight as well log off for the day. (though Copilot is working :) and OpenCode)
- antfarm 5mo agoClaude 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.
- risyachka 5mo agoNow new model looks so much better though on benchmarks!