Requesting o1-preview over the API

API
1

Hello,
I got good and quick answers for my last question, so perhaps also in this question I get helpful answers.

I have wrote that I analyse the grammar LLM models produce in their responses. I want to include the new model o1-preview in my analysis, but it is not reachable as the other models like GPT4o. Has anyone requested these models already over the API? Do I need a special account for that? What is with the limits? 30 requests per week are not enough for my analysis. I need at minimum 1000 requests in total.

2 Likes
2

Hi again :slight_smile:

If you’d like to access o1 models via the API, then currently you must be of Usage Tier 3 or higher. You can read up on the requirements to achieve this here.

For future reference, this page currently has the exact breakdown as to o1 model availability via API as well as via ChatGPT.

The weekly request limit only applies to the use of these models via ChatGPT. In the case of the API, rate limits apply which are significantly higher, e.g. for Tier 3 you can make 5,000 requests to o1-preview per minute.

6 Likes
3

Thanks again. Was helpful.

2 Likes
4

Ok. I am only tier 1. It seems that I have not used more than 50$ yet. I saw that tier 3 needs 100$ spend. How can I reach tier 3 quickly, to continue with my analysis. Its not about the money, if I can reach tier 3 with the mentioned 100$ it would be fine. I read that the 100$ must be spend on the API. It seems that it is not enough to have the money solely on my account (it must be spend).

5

That is incorrect (unless OpenAI has made life tough with undocumented behavior). Usage should not be a factor.

The tier level is more about how much has been paid (spent) to purchase credits or pay bills in the past, and how much time has elapsed when you make the newest payment.

You can see a bunch of forum post of people not getting tier elevations, though, OpenAI saying “fixed”, and then the problem continuing. That may give a misconception of the requirements.

It is also possible that this model can’t be “attained” by becoming the current tier level, but as it rolls out, access is switched on for those already at the tier. So I wouldn’t pay more than credits you’d use shortly just to hope to attain access.

6

Thanks for this reply, I was having trouble accessing through Tier 1.

It looks like the rate limits page has not been updated since the currently published information at the link provided. o1 models are not shown anywhere in the chart for Tier 3, or any other tier. Here is the current Tier 3 rate limits (which does not show the o1 models).

|MODEL|RPM|TPM|BATCH QUEUE LIMIT|
|---|---|---|---|
|gpt-4o|5,000|800,000|50,000,000|
|gpt-4o-mini|5,000|4,000,000|40,000,000|
|gpt-4o-realtime-preview|5,000|80,000|-|
|gpt-4-turbo|5,000|600,000|40,000,000|
|gpt-4|5,000|80,000|5,000,000|
|gpt-3.5-turbo|3,500|4,000,000|100,000,000|
|omni-moderation-*|1,000|50,000|-|
|text-embedding-3-large|5,000|5,000,000|100,000,000|
|text-embedding-3-small|5,000|5,000,000|100,000,000|
|text-embedding-ada-002|5,000|5,000,000|100,000,000|
|omni-moderation-*|500|20,000|-|
|whisper-1|5,000|-|-|
|tts-1|5,000|-|-|
|tts-1-hd|5,000|-|-|
|dall-e-2|5,000 img/min|-|-|
|dall-e-3|5,000 img/min|-|-|

This is what was published in the other mentioned link

Only Usage Tier 3, 4 and 5 API accounts can access the o1-preview and o1-mini API models:
 
Tier 5: Requires 30+ days of payment history and at least $1,000 spent on the API.
Tier 4: Requires 14+ days of payment history and at least $250 spent. 
Tier 3: Requires 7+ days of payment history and at atleast $100 spent
​
Learn more about your API usage tiers.
 
OpenAI o1-preview and o1-mini API models are subject to the following rate limits:
​
Tier 5:
o1-preview: 10,000 requests per minute
o1-mini: 30,000 requests per minute
 
Tier 4:
o1-preview: 10,000 requests per minute
o1-mini: 10,000 requests per minute
 
Tier 3:
o1-preview: 5,000 requests per minute
o1-mini: 5,000 requests per minute

There are no token limits for the o1 models. Does anyone know what they are? @jr.2509 perhaps you’ve seen it somewhere?

2 Likes
7

I suspect that the main reason why OpenAI has not updated the formal rate limit overview yet is that access to o1 models and the applicable rate limits are still frequently changing (there were three changes in the last ~3 weeks).

In terms of token limits the following applies:

image

Source: https://platform.openai.com/docs/models/o1

I hope this helps!

2 Likes
8

Developers who qualify for API usage Tier 3, 4 and 5, can prototype o1-preview and o1-mini…

When you go to organization limits, you will be shown your own account’s limits for models that you have access to. It has your tier level at the top. Models that you cannot access are not shown there even when expanded.

If you see tier 3+, do not see o1 models in the rate list, do not see o1 models in the API models endpoint listing, and are rejected when making a conformant chat completions call, then the account page also has a “help” button.

1 Like
9

Wait a sec, is o1 models already available for Tire 3? Cause I’m pretty sure that I’m at Tire 3 and no o1 models show at my playground

image
image1878Ă—838 78.1 KB

10

Yes, technically it should be available. Have you tried making an API call?

11

Yes and I’ve been rejected.

12

Screenshot 2024-10-21 at 7.44.04 PM
Screenshot 2024-10-21 at 7.44.04 PM2136Ă—1286 257 KB

it is available for tier 3,4 and 5 now check this link for more info https://help.openai.com/en/articles/9824962-openai-o1-preview-and-o1-mini-usage-limits-on-chatgpt-and-the-api

13

I am tier 4 (located in Denmark). I can only access via playground. There is no access through assistant or in the API