Gpt-4o-2024-08-06 is already available in the API

General
1

A new version of GPT-4o has been released. In the Chatbot arena, it is better than Claude 3.5 Sonnet in coding and takes first place. It is also now available through the API.

image
image1956×1174 270 KB

6 Likes
2

It looks like I can add this model through the settings page in Cursor, but it would be great if it could also be integrated into the long-context chat.

5 Likes
3

would be great to hear from devs

4

It seems it’s possible to use in Long Context Mode already when adding chatgpt-4o-latest-128k to the models and then selecting it in the Long Context Chat!

Same seems to be true for gpt-4o-2024-08-06-128k.

I’m wondering which one is better. The arena only lists chatgpt-4o-latest atm, and it surpasses Claude Sonnet 3.5 in coding by 30 points, which seems very impressive.

However, this benchmark still lists Sonnet as the winner:
https://livebench.ai/

5

are you using pro cursor or your API? It doesn’t work for me on PRO plan

6

I switched to API to test these models.

7

** Update: **

It works for me if I use this model chatgpt-4o-latest and enter my OpenAI API Key.

new_model_test_p5
new_model_test_p51507×816 52.8 KB

** Original Post: **

I hadn’t added my OpenAI API key or a custom model before.

But just now, to test, I added:

chatgpt-4o-latest-128k

(I also tried gpt-4o-latest and got the same behaviour)

And I created a secret (with ‘All Permissions’) at:

https://platform.openai.com/api-keys

And I added the secret to Cursor Settings > Models > OpenAI API Key.

I started a new chat (Ctrl + L) in Long Chat Mode and asked:

hello - can you please tell me what model you are using?

I get a message that says:

The model chatgpt-4o-latest-128k does not work with your current plan or api key

Screenshots below show:

  • Attempt after adding model, without OpenAI API Key
  • Attempt after adding model, with OpenAI API Key
  • Developer Tools Console when testing gpt-4o-latest

new_model_test_p1
new_model_test_p11505×604 41.8 KB

new_model_test_p2
new_model_test_p2824×945 37.1 KB

new_model_test_p3
new_model_test_p31502×821 46.8 KB

new_model_test_p4
new_model_test_p41910×296 27.6 KB

3 Likes
8

It appears that the model used on Livebench AI is ‘chatgpt-4o-latest ,’ which seems to be specifically designed for the ChatGPT interface rather than API calls.

EDIT: Unfortunately, it looks like the issue persists with both model names in long context chat. Still waiting on the development team for a resolution.

9

Hey @litecode

Could you try entering the name “gpt-4o-2024-08-06”? This should work for now, but not for long contexts.

10

Hi @deanrie ,

Yes I can confirm the following models work when I have entered my OpenAI API Key and added the additional models in Cursor Settings > Models.

  • chatgpt-4o-latest
  • gpt-4o-2024-08-06 (this one seems to work without OpenAI API Key)

When I select them in Ctrl + L in Normal Chat mode, they both work :+1:.

Interestingly, the model names which i tried earlier that didn’t work, still show up in my usage at https://www.cursor.com/settings:

new_model_test_p6
new_model_test_p6915×748 64.1 KB

new_model_test_p7
new_model_test_p71071×329 18.6 KB

4 Likes
11

Yes, I mean in the pro version.

1 Like
12

This model doesn’t work so well with Cursor for me. I don’t know if it needs fine tuned to work with Cursor’s backend like the other built in options we get but for now I will avoid it until it’s built in. It is not making coding changes or it’s breaking the code by adding changes without removing other code it should remove. I assume it is because it’s not fine tuned and was just manually added by me.