Using GPT 4, but Assistent says its GPT 3

[Creveoolus]

Bug description

1. Use any gpt 4 model (4, 4o, 4o-mini)
2. Ask it about their full model name
3. It says that its based on gpt 3.5, sometimes gpt 3

Screenshots

Environment

• python version: Python 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)] on win32
• location: Russia

Additional context

[Creveoolus]

[TheFirstNoob]

@Creveoolus Hi, it is normal that the model responds like this because new models are trained using the layering method. In fact, the model has new capabilities and an updated database and works as stated, namely GPT-4 + "o" version

[depis13]

I think openai-sentinel-turnstile-token header is not generated correctly. That's why only gpt3 is used now.

[hlohaus]

@depis13

I can't confirm that there is an issue with the models in the OpenAIChat provider. It’s possible that the language model is providing inaccurate information (sometimes referred to as “hallucinating”) or lacks sufficient knowledge on the topic However, when I use the O1 model, the response format includes clear reasoning, which aligns with the expected behavior of the O1 model.

[depis13]

@hlohaus I don't use this repository, but I used the developments on generating authentication headers in my project. In my project, I noticed that I can no longer receive links from chatgpt. In the chatgpt gui, it looks like the attached screenshot. The first request is made using the library, the second request is made using a browser without automation. It is possible, of course, that I transferred the developments from this repository to my project not very correctly, but up to a certain point I was able to get links using my code.

[hlohaus]

The issue is not with gpt4free; our responses include embedded links and a summarized list at the conclusion. It's possible your service provider is filtering these links from the response. @depis13