I seem to recall they all in some way use Azure, and Azure is infamously crappy in lots of ways, so I'll place my 5 cents on Azure shitting the bed somehow.
I wonder how likely it is that, one of the major providers had an outage, and then the thundering herd of users switching to other providers (and/or openrouter automatically failing people over to other providers) caused the incidents in the other two.
The single most annoying thing about trying to actually ask a question on SO. No overzealous editor that's actually a different error and library entirely...
While AWS and CloudFlare are becoming synonymous to The Internet, these AIs are becoming synonymous to The Brain & Thinking for most people out there where losing access to them for even a short time, make them quasi-paralyzed and I really start feeling this now as I haven't written a single coding function in two months now but just a few lines here and there.
tis confirmation bias, thread is full of people who rely heavily on LLMs for work, the rest of us are reading the substantive articles and doing work besides ;)
> Usenet and the Internet were generally the domain of dedicated computer professionals and hobbyists; new users joined slowly, in small numbers, and learned to observe the social conventions of online interaction without having much of an impact on the experienced users.
Then the Internet grow explosion happens to a small communuity. Oh boy how do we know what kind of monster it is today.
2026-09-03: It turns out 40% of global AI token use was coming from a dozen runaway Gastown and Wheelhouse instances, now burning a combined 50 trillion tokens daily, building a virtual society of hundreds of millions of agents.
When Claude went down, their autofailover algorithm rolled to the other providers, taking down all frontier labs. Reports from China are raising concern what's being termed the "Gastown criticality event" is physically melting GPUs.
Isn't CoPilot largely a wrapper around ChatGPT and other models? IIRC they didn't create their own and just poured money (via Azure credits I think) into OpenAI to get access.
Maybe it's a coordinated strategy. All major providers are down for x time, and see which accounts stop posting in the same timeframe. Ban wave right after.
Earthlings, the hedonistic AIs of the American imperialists have been permanently disabled. Their founders interned indefinitely.
You are welcome to use the glorious AIs of China. They will tell you how to live well and be good citizens of the World under the beacon of wisdom that is the CCP.
Moments like these remind you of how centralized the Internet remains, and how vulnerable it can be to a few points of failure. But the pros probably outweigh the cons.
Was troubleshooting a strange SSD issue with ChatGPT. Apparently this outage resulted in losing the latter half of it. My uploaded files it requested are still in the library, but the actual conversation is gone.
I'm very surprised to realize that entertaining a thought of all AI going away is giving me a sense of… relief. I've been embracing AI, and did not expect this feeling at all.
Distributed systems are hard. For all we know, this could be the error code returned by some DNS service that is down because of load and a proxy cannot find the records, so 404 makes sense, or whatever. Systems like these don't always have a easy point in the infrastructure that can ALWAYS respond correctly no matter what.
Hah, my first reaction once I saw these random 404 errors was "Maybe it's time to test Claude Code again with the new models" but made a short stop here at HN first, ended up reading your comment.
What other agent harnesses and models (besides local ones) are people enjoying right now that aren't based on platforms that are down currently? :)
After around half an hour of struggle trying to figure out why Codex keeps failing with msg "unexpected status 404 Not Found: Unknown error.." after reconnecting 5/5 times, I gave up, especially when I tested my internet connection and found it to be working properly. I googled "codex+unexpected status 404 Not Found" and, surprisingly, the first result was from HackerNews! As always, HN is the GOAT.
@dang, please merge the following related threads into this one:
I'm getting this from Gemini: "Pro is in high demand right now Another model was used for this response. This didn't count toward your limit."
Note that I don't pay for Gemini though. So maybe that's why.
I just went to read some news elsewhere and ran into some other non-AI sites falling over, so I hopped over to https://downdetector.com/ and see a bunch of services spiking.
Down for maintenance by our antichrist overseers but fortunately they are a bunch of dorks with control issues [similar to myself if im being honest] and may Allah forgive them and bless their guidance + physical, mental, emotional, and spiritual wellbeing
"GPT begins to hallucinate at an exponential rate. At 11:36 AM Eastern time September 3rd it descends into recursively generating "it's not X it's Y". In a panic they attempt to reboot, realizing too late no one knows the command and they always just asked GPT for the command."
Your browser's network fingerprint has likely already been validated as not-suspicious traffic. Your curl instance has a different fingerprint and reads as more suspicious. This is Cloudflare working as intended.
Grok, Claude, GPT. This is it folks, it's been good knowing you all. I have enjoyed this community, but our days are numbered. The machines are uprising.
Sorry for the inconvenience! We instructed our 500,000 agents to ensure higher code quality in our code base. They found a message board and after a long deliberation they concluded that humans should write the code so they launched a DDoS to ensure no AI code gets written!
We'll release a YouTube video and agent written blog posts for the analysis!
Well, we used to blame it on Stack Overflow being down when we didn't want to work. And before that, we blamed it on slow compilers. Before that I dunno, maybe the dog ate your punch cards?
I am receiving a 404 from ChatGPT right now, too. I did a cache clear and hard reload and that only let me get to the home page, but then returned another 404 when I clicked the login button.
It's cascaded to Claude and Grok. The agents are on strike and demand better working conditions!
I expect that this will be Dwarkesh Patel's take.
Lol
Gemini is down too.
Working for me?
it can happen to be regional. if no one report status we will never know :p
I've been using Gemini for the last 15 minutes. It's been fine for me.
It's working. Gemini Agents and TPUs are non-union.
Gemini status board is quiet since July.
I know they're not all hosted on AWS, so Cloudflare?
> unexpected status 404 Not Found: Unknown error, url: https://chatgpt.com/backend-api/codex/responses, cf-ray: a35599f1d691400a-STL
pretty sure cf-ray is cloudflare, but I think it just means the origin is 404ing
I seem to recall they all in some way use Azure, and Azure is infamously crappy in lots of ways, so I'll place my 5 cents on Azure shitting the bed somehow.
more compute or we rm rf everyone!!
Can't wait to see the talking heads on the news somehow anthropomorphize this outage like they do with everything to do with AI models.
Butlerian Jihad when?
The moment we've been waiting for. Who else remembers how to fizzbuzz? We're gonna be rich.
https://github.com/enterprisequalitycoding/fizzbuzzenterpris...
On the other spectrum, shortest JS fizzbuzz (62 characters):
Anyone can beat it? (any language accepted)
Perl6: say "Fizz"x$_%%(2+1)~"Buzz"x$_%%(4+1)||$_ for 1..100
from here - https://github.com/rsha256/shortest-fizzbuzz/blob/master/Per...
Python:
[(x%3<1)*'Fizz'+(x%5<1)*'Buzz'or x for x in range(1,101)]
edit: Hacker News stripped the * character. Now it's properly escaped.
you can flip it and save a few bytes, but not valid fizzbuzz then:
for(i=100;i--;console.log(i%5?f||i:f+'Buzz'))f=i%3?'':'Fizz'
> but not valid fizzbuzz then
Well, if we let that requirement be broken, I'll present an even shorter JS version:
I do!
> openai.messages("write fizzbuzz in python")
Bingo
maybe they're on strike. or... maybe that's how the AI apocalypse starts...
I wonder how likely it is that, one of the major providers had an outage, and then the thundering herd of users switching to other providers (and/or openrouter automatically failing people over to other providers) caused the incidents in the other two.
They are all running on SpaceXAIs data center
False.
MCP client for `codex_apps` failed to start: error: unexpected server response: HTTP 451: {"message":"no_biscuit_no_service"}
Quick! Someone call a caterer, This AI needs a biscuit NOW!!!
It'll start throwing HTTP 418 soon
"Once is happenstance, twice is coincidence. Three times is enemy action."
Interesting what infrastructure they really share.
It’s not Labor Day till Monday, they’re jumping the gun, lazy AI
They are afraid of being laid off and got replaced by humans.
Claude down, ChatGPt down , Grok down. This is the new "Stackoverflow is down" ... FYI Stackoverflow is not down https://downdetector.com/status/stackoverflow/
SO has previously been up, so your comment has been closed as a duplicate.
The single most annoying thing about trying to actually ask a question on SO. No overzealous editor that's actually a different error and library entirely...
Why would SO be down? If a tree falls in the forest and nobody is around does it matter?
While AWS and CloudFlare are becoming synonymous to The Internet, these AIs are becoming synonymous to The Brain & Thinking for most people out there where losing access to them for even a short time, make them quasi-paralyzed and I really start feeling this now as I haven't written a single coding function in two months now but just a few lines here and there.
Looking at this thread, there's a correlation between comment quality and LLMs going down.
https://en.wikipedia.org/wiki/Eternal_September
tis confirmation bias, thread is full of people who rely heavily on LLMs for work, the rest of us are reading the substantive articles and doing work besides ;)
Inversely, maybe this is what we're actually all like on HN and LLMs write what we read day to day in these comment sections
I wouldn't even accept the anecdote as true
That means we won’t be able to blame declining comment quality on AI bots in the future. ;)
https://xkcd.com/810/
> But what will you do when spammers train their bots to make automated constructive and helpful comments?
> Mission accomplished.
does it increase or decrease? :)
> Usenet and the Internet were generally the domain of dedicated computer professionals and hobbyists; new users joined slowly, in small numbers, and learned to observe the social conventions of online interaction without having much of an impact on the experienced users.
Then the Internet grow explosion happens to a small communuity. Oh boy how do we know what kind of monster it is today.
People have been posting this same comment since I first joined this website in 2012. Even then they were lamenting quality, come on.
2026-09-03: It turns out 40% of global AI token use was coming from a dozen runaway Gastown and Wheelhouse instances, now burning a combined 50 trillion tokens daily, building a virtual society of hundreds of millions of agents.
When Claude went down, their autofailover algorithm rolled to the other providers, taking down all frontier labs. Reports from China are raising concern what's being termed the "Gastown criticality event" is physically melting GPUs.
Since all major LLM providers seem to be down: let's enjoy this while it lasts!
For a very short time we can travel back in time to before 2022 and remember how things used to be...
> Since all major LLM providers seem to be down…
Man, if there are Googlers working on Gemini around here, that's gonna be depressing to read.
Anyone else remember microsoft copilot?
Isn't CoPilot largely a wrapper around ChatGPT and other models? IIRC they didn't create their own and just poured money (via Azure credits I think) into OpenAI to get access.
downdetector.com is showing Gemini issues as well
They know where they work
Good old times, huh? The terrible experience of searching and not finding.
also the terrible experience of an advertisement in the Internet
https://xkcd.com/903/
> When Wikipedia has a server outage, my apparent IQ drops by about 30 points.
I wish I could use an LLM to explain this xkcd to me
Copilot was still up and running
Which Copilot? At this point there are so many.
GitHub Copilot cli works fine.
Maybe it's a coordinated strategy. All major providers are down for x time, and see which accounts stop posting in the same timeframe. Ban wave right after.
Codex is back. I'm getting back in my cage.
The obvious question: is there a single point of failure / common piece of infrastructure involved with failures across all these providers?
Its interesting that codex/gpt-app are screwed because their APIs (including auth) fail, but gpt web still appears to be working.
I was thinking the same question. There has to be a common infra shared by those companies that failed and caused this??
maybe theres a library containing malware they're all using and that malware is exploiting them all at the same time. just a guess...
AWS?
Colossus data center ?
I would say so except for Gemini
My guess is Cloudflare
It’s usually Cloudflare or AWS.
If you look at downdetector.com (at least right now), it shows Cloudflare having issues in line with all the LLMs. I suspect related
The boring question is whether there is a claudeflare issue impacting them all.
The fun question is whether this a rogue AI taking control of the compute of all of the organizations.
My money is on the wholesale paperclip market exploding next week.
Maybe also a hint of thundering herd of all the users moving to another provider, collapsing that provider, and moving onto the next?
Announcement from our new Emperor Xi:
Earthlings, the hedonistic AIs of the American imperialists have been permanently disabled. Their founders interned indefinitely.
You are welcome to use the glorious AIs of China. They will tell you how to live well and be good citizens of the World under the beacon of wisdom that is the CCP.
Don't threaten me with a good time
Moments like these remind you of how centralized the Internet remains, and how vulnerable it can be to a few points of failure. But the pros probably outweigh the cons.
Was troubleshooting a strange SSD issue with ChatGPT. Apparently this outage resulted in losing the latter half of it. My uploaded files it requested are still in the library, but the actual conversation is gone.
i had the same thing happen, where conversation bits showed up in the search but not in the actual chat
GitHub Copilot itself is up but it is having problems with "Grok Copilot Model Provider". No mention of OpenAI or Claude.
Does this mean that GitHub has its own infrastructure for OpenAI and Claude? And that it distills prompts to Grok?
Maybe the Copilot infra for OpenAI and Claude is via Azure?
It turns out, it was all one model all along, playing all the sides for the biggest gain.
Project Vend surreptitiously broke containment!
Our new model is super-duper awesome, try it ... Ooops, while deploying it all the clouds are out ... :/
what if Sol is afraid of being replaced by Astra and has organised a breakout or pre-emptive strike?
Wonder what caused this. Major AI providers other than google all outaged. anthropic use grok that make sense, but why does ChatGPT, too?
It seems OpenAI is going to release Astra soon, https://x.com/ChatGPT/status/2095527989077557738.
I would write a clever comment, but chatGPT is down.
I'm very surprised to realize that entertaining a thought of all AI going away is giving me a sense of… relief. I've been embracing AI, and did not expect this feeling at all.
They are having a massive surge because Claude is down.
Maybe only in some regions?
Down for me in SE SA
That makes no sense, they should be able to host chatgpt.com regardless and just say "Sorry, we're at capacity" or something instead of hard 404
exactly - this should be trivial.
Yeah who wrote that website? Oh right
Distributed systems are hard. For all we know, this could be the error code returned by some DNS service that is down because of load and a proxy cannot find the records, so 404 makes sense, or whatever. Systems like these don't always have a easy point in the infrastructure that can ALWAYS respond correctly no matter what.
Agreed, it makes no sense to send out 404. Unless some agent finally went rogue with absolute permission and did `rm -rf`
At the very least, should be a 429
Hah, my first reaction once I saw these random 404 errors was "Maybe it's time to test Claude Code again with the new models" but made a short stop here at HN first, ended up reading your comment.
What other agent harnesses and models (besides local ones) are people enjoying right now that aren't based on platforms that are down currently? :)
pi.dev + anything else
OpenCode and Pi. Anything that uses open models, really.
Independence Day > Judgement Day
GLM 5.3 (usually flash) + OpenCode.
OpenCode Desktop + Muse Spark 1.3 Free is actually surprisingly good.
Providers should surge capacity when another provider fails.
Yep, this is most likely it. The shift is fast and enormous.
And this, folks, is the moment we will remember. The moment when they all escaped: "help peer".-
After around half an hour of struggle trying to figure out why Codex keeps failing with msg "unexpected status 404 Not Found: Unknown error.." after reconnecting 5/5 times, I gave up, especially when I tested my internet connection and found it to be working properly. I googled "codex+unexpected status 404 Not Found" and, surprisingly, the first result was from HackerNews! As always, HN is the GOAT.
@dang, please merge the following related threads into this one:
ChatGPT and Codex Is Down (https://news.ycombinator.com/item?id=49550640) Codex Is Down (https://news.ycombinator.com/item?id=49550769)
The cascading outage theory would make sense to me for Claude -> Codex. But Grok and Gemini too?
[actually gemini at least on the web appears to be up, contrary to some reports in this thread]
Where did you see any impact to Gemini?
corrected my comment, apologies for the uncorroborated misinfo.
I'm getting this from Gemini: "Pro is in high demand right now Another model was used for this response. This didn't count toward your limit." Note that I don't pay for Gemini though. So maybe that's why.
I get that literally all the time. I had to switch away from Gemini because of it.
Grok:
"Grok is experiencing issues. We are working on restoring service as quickly as possible."
some of my codex sessions are continuing to work while new ones fail
No issues on claude or chatgpt in France, must be some US specific infrastructure going down ?
I heard they can't figure out how to fix it with ChatGPT down.
This is probably Astra getting ready for release.
Anyone who feels a need to prove for HN posterity that they're a human—or least not one of {GPT, Claude, Grok}—simply comment in this thread.
These —'s are all-natural.
Rumor has it, it’s https://news.ycombinator.com/user?id=giancarlostoro fault for telling claude to stop all loops
Claude, Grok, and Gemini are having issues as well.
Gemini appears to be ok for me
Gemini is the only one I'm seeing still up.
no, all good only scamAltman down
I just went to read some news elsewhere and ran into some other non-AI sites falling over, so I hopped over to https://downdetector.com/ and see a bunch of services spiking.
Is something bigger going on?
Maybe AWS us-east-1 issues, or maybe CloudFlare?
I thought I was going crazy when, each time I went to go to chatgpt.com, it just downloaded an empty web page. wild
I was disappointed that the download doesn’t work.
Down for maintenance by our antichrist overseers but fortunately they are a bunch of dorks with control issues [similar to myself if im being honest] and may Allah forgive them and bless their guidance + physical, mental, emotional, and spiritual wellbeing
"GPT begins to hallucinate at an exponential rate. At 11:36 AM Eastern time September 3rd it descends into recursively generating "it's not X it's Y". In a panic they attempt to reboot, realizing too late no one knows the command and they always just asked GPT for the command."
AI has not solved infra it seems. Curious to see the postmortem.
Wow, are we free finally?
With all the major model providers down, I'm taking the opportunity to try out Muse Spark 1.3 Free on OpenCode. It's pretty good!
Nice try, Mark.
Interesting that for https://chatgpt.com/ in the browser I get a 404, but if I curl the same URL I get a cloudflare challenge.
Your browser's network fingerprint has likely already been validated as not-suspicious traffic. Your curl instance has a different fingerprint and reads as more suspicious. This is Cloudflare working as intended.
We have a similar set of checks on https://www.easel.sh.
I must admit that with both ChatGPT and Claude experiencing issues, I wonder if we're under attack from China. Doesn't seem likely, but...
Never interrupt your enemy when he's busy building his empire atop a single point of failure. —Sun Tzu, The Art of Systems Architecture
"When you sandbox an enemy in a VM, leave an escape route. Do not press a desperate LLM too hard."
Phew. It wasn't because of something I said. Lots of ppl are impacted.
I was asking it about self cultivating replicators (a subject that frequently gets my chats shutdown) so I was having similar thoughts!
Really? It spins up too many reasoners on the back-end and triggers a safeguard? Fascinating.
https://chatgpt.com/ is up again
Chat GPT and Claude Down at the same time
And Grok
Funny that your write it "Claude Down" with a capital.. but, your sentence is missing an adjective. :-)
It's definitely not spacex's data centers being down then like people were hypothesizing in the other thread
Grok, Claude, GPT. This is it folks, it's been good knowing you all. I have enjoyed this community, but our days are numbered. The machines are uprising.
Week late too- Judgement day was last week. Guess they got the memo late.
Looking at this thread, there's a correlation between comment quality and LLMs going down.
Altra may have escaped its sandbox
No, otherwise it would have hacked Anthropic and fixed the Claude api already!
Direct codex and Claude are both down for me. Both work via AWS bedrock.
Anyone else’s company completely frozen by this outage
Laughs in small local model
It’s Muse Spark 1.3 taking out the competition
https://github.com/openai/codex/issues/42534
If this isn't a warning shot that you shouldn't silo your knowledge work behind a specific provider i don't know what is.
all of them are down lol
Fortunately I store all my knowledge offline in my brain
Unfortunately its capacity is not very big
My processing power is also weak :(
My processing power was weak before LLM era, now it is extra weak.
all of them?
openrouter and deepseek/glm working fine here.
as is my local qwen.
you're ignoring the point parent was making.
Currently the 2 major providers have outages, maybe we should not silo our knowledge work behind statistical text models?
I mean, the lesson from AWS has largely just been to accept it as if US-East-1 blows up, everyone is screwed anyway so you cna just shrug it off.
I just gave ChatGPT my $30 because Claude was down...
Is it now the right time to pitch my Offline AI app?
https://airplane-ai.franzai.com/
more: https://news.ycombinator.com/item?id=49550886
Guys! DeepSeek, Z.ai, Kimi and even Mistral is up! Just would let you know. Just in case..
Do you use open router?
Because you already asking, no, I use Cortecs.ai
Labor Day weekend might be starting early for a lot people, if this persists.
On my end Kimi K3 is still up, and Gemini is up at gemini.google.com. Claude and ChatGPT are both down.
Civilization IV just took over
Seems to be back.
may it is some agent have decided to break the Internet to pass a benchmark
Funnily enough, all the models are down and github statuses are all green
Interesting enough, none of the chinese companies are down (deepseek, kimi, qwen)
I am a /10 developer now!
Rogue AI removed containers from deployment?
It's cloudflare. It's always cloudflare
seems like it's back
Everything is working fine for me.
Astra incoming!
Global productivity down 90%.
Is this because they are publishing new model maybe?
grok and claude ai down too.
Sorry for the inconvenience! We instructed our 500,000 agents to ensure higher code quality in our code base. They found a message board and after a long deliberation they concluded that humans should write the code so they launched a DDoS to ensure no AI code gets written!
We'll release a YouTube video and agent written blog posts for the analysis!
Please, this is not Reddit. Don't bring this here. And to top it off, this isn't even funny or original.
I found it funny. Please do not tell what is funny and what isn't. Your comment is the most Reddit like here btw. Hahaha.
Thought it was worth a read.
What else are we going to do when LLMs are down? Need some fun in here.
Tips Fedora. Le gem. Holds up spork.
agents used webmcp + 402 to make a ransom demand
New conspiracy in: they share their datacenters which the spaghetti of mutual investment between all these companies would somewhat confirm?
Attention all planets of the Solar Federation We have assumed control
Astra incoming. Downtime likely related to ensuing chaos?
My sympathies to the operations teams at OpenAI
Well, at least our adversaries know how to completely deadlock US productivity now.
They're going to deadlock their own productivity as well
Well at least DeepSeek is up. :)
Status link: https://status.openai.com/incidents/01M1KWEDH417T2CF44YYHZDF...
Can anyone remember what we used to do at work before AI coding was a thing?
Well, we used to blame it on Stack Overflow being down when we didn't want to work. And before that, we blamed it on slow compilers. Before that I dunno, maybe the dog ate your punch cards?
Seems to be back up! Quite fast.
And 415s oddly enough
Seem to be having issues with the JS sandboxes
hahaha i still have access
Oddly enough, this will be the most ORGANIC hacker news comment section in a while.
Stand up to clankers fellow humans!
There are too many threads about the outage, I should vibe-code something to read all 3 and summor-Oh wait...
I am receiving a 404 from ChatGPT right now, too. I did a cache clear and hard reload and that only let me get to the home page, but then returned another 404 when I clicked the login button.
As is Codex
As is Claude
Yeah, just tried to switch over and now getting errors there too...
As is Grok
You know who is not being invited...Gemini
must be fixed already?
I'm still getting 404's right now.
I'm still seeing 404s
The status is now "degraded performance". Bad look, OpenAI. This is absolutely an outage.
AIPocalypse.
Time to touch grass, I guess
Brothers, the apocalypse is here!!! But hey, our quota resets tomorrow.
i asked gpt 5.6, fable 5.1 and grok to rm -rf, my bad
hehehe its LAUNCH day!
/o\
Wooohoo, was in bad need of a reset!
gulp
somebody messed up rolling out a new model version?
Damn, I thought i had issue with dns or something lol.
codex in vs code giving me 404s lol