What?

https://openai.com/index/openai-o3-mini/

Why is this interesting?

• Arc-AGI challenge: Test AI’s ability to handle completely new tasks ("North Star towards AGI”)
• Reasoning models represent new scaling paradigm

https://x.com/mckaywrigley/status/1885426569487040745

https://www.reddit.com/r/OpenAI/comments/1iemnvi/comment/ma94r0z/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

Been playing around with it for a bit. Seems roughly on par with o1 for my use cases. Overall pretty sweet deal for free users. Big jump from 4o for certain tasks.

Access?

# Model
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="o3-mini") 

How To Use?

• Some confusion on o1 models
• Give it briefs (c/o @benhylak)

https://www.latent.space/p/o1-skill-issue

• Focus on WHAT (you want), not HOW (you want it to think, roleplay)
  • Avoid chain-of-thought
  • Give it a defined end-state

Use-cases