Hello!
I just wanted to ask if the o3-mini is currently a better option for a chatbot than the 4o.
I’m asking this because of the big difference in price per token, as the o3-mini is literally less than half the price.
I really appreciate your response since, if the difference is minimal (or if I even hear that the o3-mini is actually better), I will definitely go for the o3-mini because of its price.
P.S.: I’m using an Assistant for the chatbot, and I understand that o3-mini is currently not available for the Assistant.
What purpose will your chatbot fulfill? o3-mini should have stronger reasoning skills while 4o should have broader world knowledge. If you only need it to say “Hello” and have conversations of no huge importance, you might even choose the very cheap 4o-mini.
Nope, o3-mini is available for assistants today!
Although, what type of chatbot? To @civilianemail’s point, GPT-4o is best for generating fast text and chatting… o3-mini is great at logic, reasoning, etc.
3 Likes
@edwinarbus @civilianemail Oh, great news! However, I think that, as both of you said, I won’t need the o3-mini.
My chatbot is for an enterprise that sells products, so I have a lot of instructions in the “system instruction.” When I tested it with the 4o-mini, it worked, but it didn’t follow some specific orders.
Unrelated to the main question, but do you think I should create an assistant for each task and have another assistant decide which one to use at each moment? Or would a single assistant handling everything be fine?
I’m asking because I didn’t quite understand whether the system instructions count as input tokens, and also because I don’t want the assistant to get overwhelmed with too much text.
On the face value, O3-mini is faster and cheaper than 4o.
However, O3-mini is reasoning model, it means you are paying not only for input and output but for the reasoning tokens as well and if your “reasoning effort” param is set to medium or high, there will be substantial reasoning tokens used.
You said you are building an enterprise level chatbot so I assume there will be rag involved across various company documents, and reasoning across these documents will cost a lot of reasoning tokens.
Due to this, it’s safe to say o3-mini’s price is on par with 4o.
So, if you want to optimize for speed, go for o3-mini.
Is 03 mini ready to be used for assistants? I do not see it in the options.
1 Like
input is 50% cached. so it might be a better if you need reasoning.
1 Like
Why would i use RAG, if the assistant has all the info in their system instructions, and can also use function calling?
From OpenAI’s news page (https://openai.com/index/openai-o3-mini/).
o3-mini is rolling out in the Chat Completions API, Assistants API, and Batch API starting today to select developers in API usage tiers 3-5.