Elevated errors on Claude.ai, API, Claude Code (claudestatus.com)

by redm 224 comments 243 points
Read article View on HN

224 comments

[−] edf13 30d ago
Seems to be a very regular occurrence starting around this time of day (14:30 UTC)...

Claude Code returning: API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"---"}

Over and over again!

[−] walthamstow 30d ago
US Pacific comes online while London is still working and they can't handle it. $380bn valuation btw.
[−] jjcm 30d ago
No amount of valuation can fix global supply issues for GPUs for inference unfortunately.

I suspect they're highly oversubscribed, thus the reason why we're seeing them do other things to cut down on inference cost (ie changing their default thinking length).

[−] natpalmer1776 30d ago
Remember when OpenAI wasn’t allowing new subscriptions to their ChatGPT pro plans because they were oversubscribed? Pepperidge Farms remembers.
[−] andai 30d ago
Wouldn't that be good? I remember back in the day you could only get Gmail thru an invite, it was an awesome strategy. "Currently closed for applications" creates FOMO. They'd just need to actually get the GPUs in relatively short supply. They could do it in bursts though, right? "Now accepting applications for a short time."

I'm not an internet marketer but that sounds like a win win to me. People feel special, they get extra hype, and the service isn't broken.

[−] hirako2000 30d ago
In the case of Gmail that was fake scarcity.

In the case of Anthropic is fake availability.

Sam Altman explained the idea is to scale the thing up, and see what happens.

He hadn't claimed to offer a solution to the supply problem that would unfold.

[−] bruckie 29d ago
Are you sure it was fake scarcity for Gmail? IIRC they did it because they were worried about systems falling over if it grew too fast, and discovered the marketing benefits as a side effect.
[−] iainmerrick 29d ago
Are you mixing up Anthropic and OpenAI here?
[−] hirako2000 29d ago
I didn't. Anthropic and others followed the concept of scaling up models and worry about efficiency and availability later. Sam likely didn't invent the idea but he talked about it.
[−] the_gipsy 30d ago
Yes, "Pepperidge farm remembers" is usually about how something used to be good.
[−] CoastalCoder 30d ago
Yeah, but there was a spoof on that (in Family Guy?). It was a tie in to the movie "I Know what you Did last Summer", IIRC.
[−] joquarky 30d ago
Google Wave demonstrated that this doesn't always work.
[−] scratchyone 30d ago
maybe, but the response to GPU shortages being increased error rates is the concern imo. they could implement queuing or delayed response times. it's been long enough that they've had plenty of time to implement things like this, at least on their web-ui where they have full control. instead it still just errors with no further information.
[−] skeledrew 30d ago
I've been experiencing a good amount of delays (says it's taking extra time to really think, etc), and I'm using during off-peak time.
[−] scratchyone 30d ago
i notice that as well. most of the time when i see those it has a retry counter also and i can see it trying and failing multiple requests haha. almost never succeeds in producing a response when i see those though, eventually just errors out completely.
[−] hirako2000 30d ago
Coding is a problem solved. Claud writes the code. I edit it. I code around it.

Engineer roles dead in 6 months.

[−] post-it 30d ago

> I edit it. I code around it.

You're never gonna guess what software engineers do.

[−] bulbar 29d ago
Because of the context I would think this is sarcasm, but I am not sure.
[−] hirako2000 29d ago
It is.
[−] zachncst 30d ago
Sure but we don't need GPUs to log in.
[−] sobellian 30d ago
Their issues seem to extend well beyond inference into services like auth.
[−] ryandrake 30d ago
Yes. Whenever these outages happen, it always seems that it's their login system that is broken.
[−] bostik 30d ago
That implies that either the auth is too heavy (possible, ish) or their systems don't degrade gracefully enough and many different types of failures propagate up and out all the way to their outermost layer, ie. auth (more plausible).

Disclosure: I have scars from a distributed system where errors propagated outwards and took down auth...

[−] AlecSchueler 29d ago

> thus the reason why we're seeing them do other things to cut down on inference cost (ie changing their default thinking length).

The dynamic thinking and response length is funny enough the best upgrade I've experienced with the service for more than a year. I really appreciate that when I say or ask something simple the answer now just comes back as a single sentence without having to manually toggle "concise" mode on and off again.

[−] paulddraper 29d ago
A. These aren’t rate limit errors from the API.

B. Everything is down, even auth.

[−] ai-x 30d ago
This precisely justifies Anthropic's market cap to be higher.
[−] dsr_ 30d ago
Demand at an unsustainably low price does not imply demand at a sustainable price.
[−] bigbadfeline 29d ago
I'm pretty sure ai-x writes sarcasm and skips the /s for pure fun. Personally, I'm amused and I like what he's doing. Others have done it before him though, it's not a new trick.
[−] tucnak 30d ago
Assuming perfectly efficient business
[−] azalemeth 30d ago
I literally just came to HN to ask if I was alone with the acurséd "API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"…"}" greeting me and telling me to get back to using my brain!
[−] xnorswap 30d ago
500-series errors are server-side, 400 series are client side.

A 500 error is almost never "just you".

( 404 is a client error, because it's the client requesting a file that does not exist, a problem with the client, not the server, who is _obviously_ blameless in the file not existing. )

[−] darkwater 30d ago

> A 500 error is almost never "just you".

I know you added the defensive "almost" but if I had a dollar each time I saw a 500 due to the session cookies being sent by the client that made the backend explode - for whatever root cause - well, I would have a fatter wallet.

[−] iainmerrick 29d ago
Depending on what you mean by "made the backend explode", that is a server error, so 500 is correct!

Bad input should be a 4xx, but if the server can't cope with it, that's still a 5xx.

[−] xnorswap 30d ago
Indeed, and also there's a special circle of hell reserved for anyone who dares change the interface on a public API, and forgets about client caching leading to invalid requests but only for one or two confused users in particular.

Bonus points if due to the way that invalid requests are rejected, they are filtered out as invalid traffic and don't even show up as a spike in the application error logs.

[−] azalemeth 30d ago
I know that in principle this is true. However, I have seen claude shadow-throttle my ipv4 address (I am behind CGNAT), in line with their "VPN" policy -- so I do not trust it, frankly.
[−] paganel 30d ago

> in line with their "VPN" policy

This is how I learn that they have a "VPN" policy. Thinking of it maybe it makes sense, that is if it's what I think it is, but seems scummy nonetheless.

[−] andyjohnson0 30d ago

> Seems to be a very regular occurrence starting around this time of day (14:30 UTC)...

8.30am on the US west coast

[−] imdoxxingme 29d ago
Probably when they're permitted to start live experiments
[−] freedomben 30d ago
Yep, daily haha. Well at least this time they aren't just silently reducing thinking on the server side, which ended up making a mess in my codebase when they did that last time. I'd rather a 500 than a silent rug-pull.
[−] JamesSwift 29d ago
I tend to notice it around 4pm EST
[−] Sol- 30d ago
Once AGI is achieved, they'll reach the fabled superhuman "two nines" of uptime.
[−] jakemoshenko 30d ago
Or the more human nine fives.
[−] sensanaty 30d ago
They seem to be struggling with even a "one 9" as-is.
[−] PunchyHamster 30d ago
but the Github way of 89.09%
[−] just_once 30d ago
I guess that's called napping
[−] dust42 30d ago
Well, at least we know by now that Mythos is a mythos.
[−] m_ke 30d ago
more like 0, once they get AGI they'll capture all value themselves or sell to highest bidders
[−] no_shadowban_9 30d ago
[dead]
[−] lbriner 30d ago
Funny that I just saw this after have "Console temporarily unavailable". I am currently at the stage that: 1) I think Claude Code is very impressive 2) I think pretty much everything else about them is terrible.

* Support really poor, raised a ticket last week and have heard nothing back at all * Separation of claude.ai accounts and console accounts is super confusing * Couldn't log into the platform since I had an old org in the process of deletion even though I was invited to a new one (had to wait 7 days!) * Payments for more API credits were broken for about a week * Claude chat has really gone to s*t unless it always was. Just getting back terrible answers to simple questions. * The desktop app is a web app pretending to be a desktop app that doesn't always know it is a desktop app so you get things like, "this will only work in the desktop app". Yes I know, this is the desktop app! "Oh sorry about that but you need to use the desktop app". * mcp integration and debugging is dreadful, just a combination of generic "an error ocurred" and sometimes nothing at all * MCP only supports OAuth for shared connectors but auth key doesn't work even with "local" servers that are not necessarily local, just the config is local.

You can put those on the health status!

[−] mesmertech 30d ago
We went from "Peak hours" meaning 2x usage plus slower to now it just does 500 error

https://mesmer.tools/random/is-it-peak-hours

[−] mchusma 30d ago
Anthropic has been avoiding the hard thing, but they just need to do SOME kind of pricing thing here to shift demand. I expect there is just no amount of tricks that can handle the few hours at peak load. They need "surge pricing".

This seems reasonable surge pricing approach to me: 1. Implement surge pricing for everyone for the peak 2 hours of the day if possible. 2 hours you can work around, 5 hours is too hard. 2. Give existing customers a one time credit for surge 3. Make sure the plans just consume credits at an accelerated rate (e.g. if on the max plan, i just get 1/2 usage during peak hours). 4. Exempt sonnet/haiku from surge (so people can keep using) 5. Make "auto" settings in claude code etc automatically adapt during surge hours, so people don't get surprises by default. 6. For the first 90 days, unofficially waive the fist $100 in surge for every user but notify them. To train users about the surge, and get them used to it without having them actually pay. 7. (I don't think they would do this but this would help) allow users to fall back to using something like GLM 5.1 or Gemma 4 automatically in outages, with a partnership to handle it. Its not ideal, but i would prefer it in "partner mode" than not. IMO they can charge like 10% on top of the partner fees for this if used in other times, but during outages or surge, partner mode is free. But 100% managed by Anthropic so users don't need to set things up and can just use the Anthropic harness.

[−] nerdright 30d ago
Funny how we all come to HN when the status page is lagging behind. HN is truly the real-time status page
[−] parthdesai 30d ago
Coding is solved
[−] pton_xd 30d ago
91% uptime for Claude Code over the last 30 days. Is that accurate?! I'm not a CC user but that seems awfully low.
[−] me551ah 30d ago
What am I supposed to do now? Copy code from stackoverflow like a caveman?
[−] jakobloekke 30d ago
Have anyone found good techniques to get a session out of Claude Code, so that I can point another tool at it and pick up there? This always seems to happen at the worst possible time, after having spent an hour getting deep into something – half finished edits across files, subagents running, etc.
[−] troupo 30d ago
In the past few weeks:

- Anthropic introduced stringent limits at peak hours. By "introduced" I mean announced it on a random dev's Xitter account

- Users suddenly started burning through all of their tokens even on trivial tasks. Anthropic never truly acknowledged it, their random devs posted "we're working on it".

- One of the workarounds was to somewhat quietly reduce default reasoning to medium

- OpenClaw and "usage through other tools" banned

- Announce "redesigned Claude Code Desktop App that lets you run many parallel sessions"

- Availability is still circling down the drain

- Dario Amodei is in continuous "trust us we have AGI coding is solved we don't need programmers just give us more money" mode now

[−] jpcfl 30d ago

> claude: "API Error: 500... check status.claude.com"

> status.claude.com: "All Systems Operational"

[−] ghc 30d ago
It seems like Claude has taken Github's place in terms of developer reaction to it being unavailable. It's like everyone forgot how they did things 18 months ago.
[−] nateguchi 30d ago
Is codex a good alternative? Or does Claude have a moat...
[−] sensanaty 30d ago
But I thought coding was solved? I guess having a single 9 of availability is something we need true AGI for, we should probably give OpenAI and Anthropic another gazillion dollars to burn through to figure this out!
[−] mr_mitm 30d ago
I went back to the $20 plan and a single prompt maxed out my quota for the five hour window within 15 minutes. I used to be able to vibe code for over an hour before. This is really annoying.
[−] nafizh 30d ago
OpenAI is very good in terms of not having as much outages as Anthropic, but almost all products except Codex and the pro model is unimpressive, anthropic has the opposite situation.
[−] anonyfox 30d ago
good advertisement now to shift the tide back to openai that just works and honestly codex with gpt 5.4 is _surprisingly good_ currently, not nerfed or forgetting half the tasks along the way so far. Opus already got worse than sonnet last weeks beyond just crazy token costs, now reliabilty goes to shit and anthropic seems like using it. Meanwhile, delightful of the codex desktop app in fact, stuff seems to "just work" elegantly with good quality.
[−] DanMcInerney 30d ago
Cross your fingers they're about to drop 4.7. 4.6 came out with a bang, now it seems all the compute bottlenecks just lead to customer frustration as they get closer to releasing next model. Balancing the books over there must be a nightmare, "Well we can piss off every single customer for a week, but we'll be able to release the next model 1 week faster"
[−] chollida1 30d ago
Shades of early twitter.

Early twitter showed the fail whale as often as it showed tweets and yet it was an unstoppable juggernaut that people kept using.

[−] ivanjermakov 30d ago
I know it's not the same, but I find it incredibly funny that my home server has better uptime than giants like GitHub and Claude.
[−] pixel_popping 30d ago
Anthropic still haven't fixed the major issue where we can't paste the token as well, and now, it starts to smell fishy, they likely did it on purpose. https://github.com/anthropics/claude-code/issues/47887
[−] garff 30d ago
Yea, it's peak time. They don't have enough compute. Why do you think they are banning external subscription use. They sell subscriptions. They don't need people to use CC. That doesn't matter. And yet - they won't have people using their service outside of CC. Something is fishy.
[−] therobots927 30d ago
Think “shrinkflation” but for tokens. Anthropic is going to absolutely suck early adopters dry.
[−] stokedbits 30d ago
Why would anyone assume a new model is dropping when their status page is showing elevated errors? Are they that sloppy that they just let their status systems report failures when they are the ones deploying new infrastructure / models / etc?
[−] bamboozled 30d ago
Basically pushed the button staying up late finishing something, didn't really factor in a Claude outage in the middle of it, here's to red eyes while I use my clumsy fingers and brain to complete the task the old fashioned way.
[−] rvz 30d ago
So after 24 hours later [0], Claude went on another lunch break? It should be on its desk at all times.

[0] https://news.ycombinator.com/item?id=47753710

[−] its_down_again 30d ago
My guess is the downtime is tied to the routines rollout https://code.claude.com/docs/en/routines
[−] bix6 30d ago
Trying to talk with support and their AI bot is failing. No support email? Joke.
[−] dkackman11 30d ago
it was me. i asked it to calculate factorials without recursion and it crashed.
[−] orzi 30d ago
Maybe Mythos DDoSed it :)

Android app is still responding but no-go on claude.ai and I can't login with email

status.claude.com has an update:

Investigating - We are seeing increased errors on Claude.ai, API, and Claude Code Apr 15, 2026 - 14:53 UTC

[−] deadbabe 30d ago
While your developers are twiddling their thumbs waiting for Claude to come back online, your competitor is using alternatives to get work done right now and advancing on their go to market timeline.
[−] sausagefeet 30d ago
I just had to upgrade my plan because I ran out of tokens because medium effort had dementia and things only worked on high. Good to know I'm getting my money's worth...
[−] filament 30d ago
You'd think the robots would be better at detecting outages.
[−] thatmf 30d ago
Looks like they redesigned the page though, so that's nice.
[−] cbg0 30d ago
I'm glad I live in Europe, I can at least use the subscription I pay for even though the quality is worse even during off-peak hours.
[−] obiefernandez 30d ago
I'm getting Cloudflare protection screens while trying to use claude.ai web app. I wonder if they were getting DDoS'd...
[−] bhu8 30d ago
Feels like an issue in their caching. First non-cached turns are sent properly but everything that is second+ turn fails.
[−] sharts 30d ago
If’s also slow AF when it’s not crashing. At this point it’s likely more cost effective to just host your own models.
[−] webXL 30d ago

> time to break out Pro C# textbook

- downdetector comment

[−] jcfrei 30d ago
A few hours ago I noticed a considerable decline in code quality. It seemed the model got downgraded so I switched to codex. Anybody else noticed this? It starts to switch from deep reasoning and trying to fully grasp architectural changes to trying to solve things on a very adhoc basis. Maybe that's just my imagination or maybe that's Anthropic trying to balance the load before being fully overloaded.
[−] christinetyip 30d ago
It's been a bit disruptive to my workflows tbh. What alternatives are people using? Sell them to me please
[−] dude250711 30d ago
Coding is solved: no more coding.
[−] etchalon 30d ago
Ugh. I swear to everything if I have to start using Codex I'm going to be so mad.
[−] zurfer 30d ago
They should just fall back to gpt 5.4. would save me the hustle to setup codex.. :P
[−] frb 30d ago
It just started again, didn't it? Nothing on the status site yet...