Tier 4 subscriber getting 429 errors, please help?

API Bugs
1

TLDR; We are a Tier 4 subscriber, and inexplicably rate limited.
We are on deadline and have lost an entire day.
Now’s your time to shine, support.

Conditions:
+ Under our token limits
+ Under our rate limits
+ Within our spend
+ Unable to get any useful information

WE’RE OVER OUR SPEND, RIGHT…?

USAGE $111.94 OF $350 / $5,000 TIER MAX / AutoRecharge is on

NO, I GUESS NOT.

WAIT! IT’S OUR TOKEN USAGE, RIGHT…?

OUR TIER ................................ TIER 4

GPT-4 TPM ALLOWED ...................  2,000,000   
GPT-4 USED .......................... 23,047,350 (2 HOUR WINDOW)
PEAK UTILIZATION 
    (IN+OUT), 15-MIN 5.5MM/15 .......   ~364,772 TPM

GPT-3.5-turbo TPM ALLOWED ........... 10,000,000   	     
GPT-3.5-turbo USED .................. 11,696,213
PEAK UTILIZATION 
    (IN+OUT), 15-MIN 1.34MM/15 ......   ~89,483 TPM


GPT-4 RPM         10,000
OUR GPT-4 RPD      8,984

NO, I GUESS NOT.

OH WAIT! IT’S OUR REQUEST RATE RIGHT…?

OUR TIER ................................ TIER 4
GPT-4 RPM ALLOWED ......................  10,000   
GPT-4 REQUESTS, TOTAL ..................   8,984 (2 HOUR WINDOW)   
GPT-3.5-turbo RPM ALLOWED ..............  10,000   
GPT-3.5-turbo REQUESTS, TOTAL ..........   5,761 (2 HOUR WINDOW)

NO, I GUESS NOT.

WHEW I CAN GET RESET DATA IN THE HEADER, RIGHT?
curl -X POST
-H “Content-Type: application/json”
-H “Authorization: Bearer sk-…”
-d ‘{“model”: “gpt-3.5-turbo”, “messages”: [{“role”: “user”, “content”: “Hello!”}], “temperature”: 0.7}’
-D -
-o /dev/null
–connect-timeout 0
–max-time 120
[openai]/v1/chat/completions

HTTP/2 429 
date: Sun, 15 Dec 2024 23:41:27 GMT
content-type: application/json; charset=utf-8
content-length: 337
vary: Origin
x-request-id: req_09fdc854ee4a00fe8716f9b690508584
strict-transport-security: max-age=31536000; includeSubDomains; preload
cf-cache-status: DYNAMIC
set-cookie: __cf_bm=VNSNb2ZI58dvuphapJ86uCxKJzPjONsjz6IHK.kYmM0-1734306087-1.0.1.1-kavQnTgSYRpWAXtshiePiWUUE2t_rQnPc1Rm_oMBSGIeRDiQoJ86xjkJ0ZMVpaLOMnsDf5kN4zBft6ZIt.hmXQ; path=/; expires=Mon, 16-Dec-24 00:11:27 GMT; domain=.api.openai.com; HttpOnly; Secure
x-content-type-options: nosniff
set-cookie: _cfuvid=ZcRYkHf5.zaFWt8np1Wmlf73xOc8XDoDm39JsPydVvo-1734306087142-0.0.1.1-604800000; path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
server: cloudflare
cf-ray: 8f2a48d47d35c940-IAD
alt-svc: h3=":443"; ma=86400

NO, I GUESS NOT.

1 Like
2

A new input might fix a ghost limit, then a little wait for it to take effect.

3

I will try this. But… Does OpenAI have ‘ghost limits’, or is that speculation? I can’t see why any company would implement that sort of thing on a platform that is designed to serve users and generate revenue.

4

Tier 5 user get the same issue, have you fix it?

5

It is just some oddity that has recently cropped up. Old budget limits used on people that had never set them or upgraded them.

6

To double check, is the API key being used from the Org with the Tier 4 limit? Sometimes an account may have more than one Org under it and this can lead to issues, if you are sure this key is from the correct Org please let me know and I’ll raise the issue.

1 Like