Embedding Model Outage: text-embedding-3-small-api-ev3 model name with all 0 values

API Bugs
,
1

Hey I think we’re experiencing an outage right now due to a server-side issue with **text-embedding-3-small
**
We’re getting strange return values with response model: text-embedding-3-small-api-ev3 and an embedding vector of size 384 (instead of 1536 as expected) with all 0 values.

Anyone else seeing this? What’s going on?

4 Likes
2

I have the same issue, it looks like they are aware of a problem on their status page. https://status.openai.com/

3 Likes
3

thanks! for what it’s worth, we’re still seeing the issues with the ednpoint contrary to the status page.

4

Experiencing the same issue. Not sure how there are not more reports of this.

5

Thanks for posting the link @Brendon_Boshell!
It looks like a fix has already been implemented.

Screenshot_20260212_180657_Firefox
Screenshot_20260212_180657_Firefox1080×986 172 KB

1 Like
6

Yes, I’m also seeing ongoing issues, around half of the responses are just vectors of zeroes.

1 Like
7

Is anyone also seeing invalid 400 errors related to “Invalid ‘input’: maximum context length is 8192 tokens.” Just started after the issues earlier.

8

The issue is not related to latency. Embeddings are being returned with 384 dimensions (all 0s), instead of 1536

2 Likes
9

3-small seems working. I made several calls with and without the dimensions parameters, with and without base64, all are 1536 dimensions and can provide a similarity score.

For 3 texts, received 99909 char body
Snippet: {
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": [
        -0.0026644536,
        -0.02758758,
        0.00036771712,
        0.0039073024,
        0.04801365,
        0.017813696,
        0.063282534,
        0.003417482,
        0.030695407,
        -0.016327344,
        -0.005967926,
        -0.034929253,
        -0.036528207,
        0.0005278236,
        -0.024614878,
        -0.0004380936,
        -0.051436763,
        0.011721906,
 
(3, 1536)
Similarity:  0.33056685
Similarity:  0.33388305
text-embedding-3-small

Existing unpacker of b64:

For 3 texts, received 24966 char body
Snippet: {
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": "jSYuu6Us4rzZ2rI5pMSAO0eoRD0q1ZE8x7CBPYIxXTvIcvs8dU2GvHYuw7teKA+9tYUVveTPAzpTYMq8yIH/uY2tUr0krT88WHmYu1P3rboTKDs8ZPWNPEe3yDz2rWU90/6vvDwsz7wqPq68U6tePUeZwLtw2h+9PDvTvPzVt7zqm0e9qqADPU2UhrwN4mA8PB3LPLtwHD1NlAY93k0ZPQ3UlzwBdiq9X2Pku/bLbby82P06MHTJuxltWjwSdIo99p8cPVOOETsT3Sa8AgzTuw08+bzfl/I8xxjjOw2XzDzk3ge9ZBOWPOqMw7x2ELs7X0XcurB7S7w7lqa8U7sdvdiAmjtTuuI7JPhTPbAgeLwSdAq9DS11POqrBrxZ8Ti9qr6LO55Rm
(3, 1536)
Similarity:  0.33061928
Similarity:  0.33393234
text-embedding-3-small
1 Like
10

It appears to only be an error with encoding_format: ‘base64’

If you change this for float it fixes the issue for now.

11

Hey all, just checking in.

Is anyone still running into this? If so, could you share a few request IDs and the timestamps if possible? That'll help us dig into the logs and see what’s going on.

2 Likes
12

errors stopped appearing for me ~12:30 pm EST yesterday

1 Like
13

2 posts were split to a new topic: Random 384 dimension embeddings from text-embedding-3-small