Looks like its down (both gpt-4o and gpt-4o-mini)
2 Likes
I don’t think so, at least gpt-4o-mini is working like a charm, I’ve just tried with my RAG application.
From the announcement article found here:
Today, GPT-4o mini supports text and vision in the API, with support for text, image, video and audio inputs and outputs coming in the future.
The wording confuses me a bit. Does this mean text, image, video, and audio I/O is currently supported, or it’s still in development? I assume it’s latter but want to confirm.
Excited to try out this release in our systems!
16383 in the playground.
1 Like
Only text and vision are currently available.
1 Like
Gotcha. Thank you for confirming!
Is the pricing page accurate in saying the vision pricing is the same between 4o and 4o-mini?
1 Like
when i got the email notification about the new model, the name made me wonder, why mini? reading thru various info from the site, i never thought that there is small models category. what does smaller tasks mean? anyway, the name reminds me of a car ad in japan which i rephrase here for gpt-4o mini: 小さなモデル、大きな能力！GPT-4o ミニ。
2 Likes
However, that is “The maximum number of tokens to generate shared between the prompt and completion. The exact limit varies by model. (One token is roughly 4 characters for standard English text)”
What is the total maximum output tokens for API? For gpt-4o it is still, as far as I know, only 4K tokens.
This is incorrect.
The maximum it can generate is reported to be 16,384 in the OP and I am seeing max_tokens of up to 16,383 in the playground.
I looks like the didn’t update the max_tokens tooltip in the playground when they changed the parameter’s meaning.
2 Likes
When using gtp-4o-mini vision capabilities, I see a 25x increase of input tokens compared to gtp-4o for the same picture. Is this a bug or intended behavior?
1 Like
Intended. The cost of reading an image is the same. The text cost is different
2 Likes
It would seem the cost of processing an image is the same, so to make up the value in tokens, the base token count is increased: see the breakdown here:
https://openai.com/api/pricing/
Is the token count artificially inflated to not cannibalize gtp-4o vision sales or is there a technical necessity for it? In practice gtp-4o-mini is still ~40% cheaper for me but performs worse, so will probably not switch for now.
1 Like
In order to truly advance AI and human society, GTP should focus on mobile apps, as mobile phones are the most frequently used devices worldwide. I hope that this GTP mini can be upgraded in an all-round way, especially in the beautification of language architecture and thinking expression, otherwise it will be replaced soon.
Your comment prompted me to check my usage… whoah!
Nice. Thank you. We will be deploying tomorrow. Love your work and grateful for the powerful tool.
Is the python library for the openai api updated?
gpt-4o-mini seems to be missing in the assistants.py file in version 1.35.15. openai-python/src/openai/resources/beta/assistants.py
Why am I not finding this model when I retrieve a list of all available models using the API? Anybody knows …?
I’ve been testing mini with our image rec. app. It’s working well so far, but what I don’t understand is the number of tokens.
A typical request requires 1,757 tokens with 4-o. The same request with mini is 18,236 tokens.
The price is reduced…but why are there so many tokens?
1 Like