403 from embeddings API despite paid account + configured limits - SOLVED using inappropriate model

API
, ,
1

I’m getting a 403 to the embeddings API, this despite:

  1. Having a paid account
  2. Creating a non-default project and configuring model limits
  3. Only one API key: a project key for my non-default project
  4. Being in the “User1” tier

This has been frustrating so I hope there’s a simple explanation for what I’m seeing.

Using the list-models endpoint, I see the following models, which is what I expect:

{
  "object": "list",
  "data": [
    {
      "id": "gpt-4o",
      "object": "model",
      "created": 1715367049,
      "owned_by": "system"
    },
    {
      "id": "gpt-4o-2024-05-13",
      "object": "model",
      "created": 1715368132,
      "owned_by": "system"
    },
    {
      "id": "gpt-4o-mini-2024-07-18",
      "object": "model",
      "created": 1721172717,
      "owned_by": "system"
    },
    {
      "id": "gpt-4o-mini",
      "object": "model",
      "created": 1721172741,
      "owned_by": "system"
    }
  ]
}

But when I attempt an embeddings call with one of the allowed models, using the same auth token/project key:
POST Body:

{
  "input": "Blah",
  "model": "gpt-4o",
}

Response is a 403:

{
  "error": {
    "message": "You are not allowed to generate embeddings from this model",
    "type": "invalid_request_error",
    "param": null,
    "code": null
  }
}
2

Welcome to the Forum!

gpt-4o is indeed not intended for embeddings - it’s a chat completions model. For generating embeddings you have the choice among the following three models, all of which can be accessed under Tier 1:

Model Description Output Dimension
text-embedding-3-large Most capable embedding model for both english and non-english tasks 3,072
text-embedding-3-small Increased performance over 2nd generation ada embedding model 1,536
text-embedding-ada-002 Most capable 2nd generation embedding model, replacing 16 first generation models 1,536

The detailed API specs for embeddings models are available here.

1 Like
3

Ah-hah! Newbie error.

:man_facepalming:

THANK YOU!

1 Like
4

No worries. Happens to the best! :wink:

5

In my defense, a 403 (and that error message) is not the most helpful error in this scenario, but that’s probably something I should have figured out.

Thanks again!

1 Like
6

I will say this API seems squirrley (technical term). After repeated calls from various clients using the same API key and model (text-embedding-3-large) to the embeddings API

Bruno gives me 200s
Curl gives me a mix of 200 and 403:

{
    "error": {
        "message": "Project `proj_qGWHlze5FmF2Xv5IISYIVlM6` does not have access to model `text-embedding-3-large`",
        "type": "invalid_request_error",
        "param": null,
        "code": "model_not_found"
    }
}

And my Go code is giving me 401s.

So it seems there’s some kind of eventual consistency happening behind the scenes.

7

Things seem to have “settled down” and back on the road to success :+1:

1 Like