search

karthink / gptel

A simple LLM client for Emacs
GNU General Public License v3.0
1.03k stars 111 forks source link

Add new model: GPT-4o #313

Closed axelknock closed 1 month ago

axelknock commented 1 month ago

OpenAI has released a new model called GPT-4o, or "GPT-4 Omni". More information here.

karthink commented 1 month ago

@axelknock thanks for the PR! I'm aware of gpt-4o but haven't added it because I'm not sure about this. As I understand, gpt-4o can respond to text queries with images, like Dall-E etc. Is this true? gptel is not set up to handle base64-encoded images in responses, so this can break things.

axelknock commented 1 month ago

Currently image generation is unavailable, and I could find no definitive answer on its future broad availability. I looked at the Dalle-3 documentation and found that the API can respond with both an image or a URL for an image, depending on the response format parameter.

If this parameter were available on the 4o response, would always sending it as URL be an option?

karthink commented 1 month ago

Okay, we'll worry about it when image generation with gpt-4o is turned on. If we can't fix it easily I can make gptel display a warning when the model is chosen.

Thanks again for the PR!