Still occurring in Europe despite being marked as resolved. There was a brief period where things stabilised, we are back to intermittent `{"type":"error","error":{"type":"overloaded_error","message":"Overloaded"}}` responses.
AWS Bedrock or Vertex AI. I guess https://openrouter.ai/anthropic/claude-3.7-sonnet/providers might be worth exploring, too, as it may provide an easier switchover in case of a failure at one of the providers (I haven't used it myself).
Anthropic's uptime is the second worst out of the most used AI APIs that I've seen. Just look at the uptime of all these providers. It's not even '3' 9s.
Gemini doesn't reveal their exact uptime figures but there have been intermittent issues with their API. [3] Overall, with Anthropic own API's they are struggling to get the basics of scalability of their systems despite billions in funding.
I'm looking forward to the post mortem of this but the importance of efficient AI inference cannot be under-estimated. Just look at Mistral as an example of handling this slightly much better than the rest. [4]
Unless you have the money, one should just consider the cloud hosted versions of Claude by AWS or GCP rather than their direct APIs, if you cannot tolerate the constant outages.
For some reason, I had subscribed to OpenAI's service updates via email and not a day goes by without some some email about some downtime.Though I am a bit surprised to see it as the best one of the lot... they might be putting their money to good use.
Basic questions: is there a rate limited 'anonymous' token to access the HTTP api?
Or do we need to create an account in order to get a secret? If yes, does this account creation is compatible with noscript/basic (x)html browsers? And if requiring an email, does it support nodns(ipv6 preferably) email addresses (coze I am self-hosted)?
EDIT 2: stable again after lots of intermittent errors.
EDIT: re-opened https://status.anthropic.com/incidents/3j3zbrx1cfbt
Still occurring in Europe despite being marked as resolved. There was a brief period where things stabilised, we are back to intermittent `{"type":"error","error":{"type":"overloaded_error","message":"Overloaded"}}` responses.
Can't confirm, from Berlin everything is available for me.
I guess I should support AWS Bedrock as an alternative in my app that uses Claude:
https://aws.amazon.com/bedrock/claude/
How many outages has Anthropic had? I have to assume AWS has had less given their pedigree.
(Btw is cursor genetic coding down as well?)
Weirdly enough OpenRouter reports issues with Claude 3.7 on Bedrock as well: https://openrouter.ai/anthropic/claude-3.7-sonnet
AWS Bedrock or Vertex AI. I guess https://openrouter.ai/anthropic/claude-3.7-sonnet/providers might be worth exploring, too, as it may provide an easier switchover in case of a failure at one of the providers (I haven't used it myself).
Works for me (I'm not in europe but my VPN is)
Anthropic's uptime is the second worst out of the most used AI APIs that I've seen. Just look at the uptime of all these providers. It's not even '3' 9s.
Gemini doesn't reveal their exact uptime figures but there have been intermittent issues with their API. [3] Overall, with Anthropic own API's they are struggling to get the basics of scalability of their systems despite billions in funding.I'm looking forward to the post mortem of this but the importance of efficient AI inference cannot be under-estimated. Just look at Mistral as an example of handling this slightly much better than the rest. [4]
Unless you have the money, one should just consider the cloud hosted versions of Claude by AWS or GCP rather than their direct APIs, if you cannot tolerate the constant outages.
[0] https://status.openai.com/
[1] https://status.anthropic.com/
[2] https://status.deepseek.com/
[3] https://aistudio.google.com/status
[4] https://status.mistral.ai/
For some reason, I had subscribed to OpenAI's service updates via email and not a day goes by without some some email about some downtime.Though I am a bit surprised to see it as the best one of the lot... they might be putting their money to good use.
Just a nitpick: OpenAI is on "3 9s" according to your own post. The nines left to the comma get counted. Source: https://en.wikipedia.org/wiki/High_availability#Percentage_c...
Basic questions: is there a rate limited 'anonymous' token to access the HTTP api?
Or do we need to create an account in order to get a secret? If yes, does this account creation is compatible with noscript/basic (x)html browsers? And if requiring an email, does it support nodns(ipv6 preferably) email addresses (coze I am self-hosted)?