If you still call o3 a pure LLM, then I'm wrong.

o3 is so much more than an LLM, the most heavylifting is done by the Test Time Reasoning based on something similar to MCTS.

You can read more here:

https://medium.com/aiguys/openai-achieved-agi-with-its-new-o3-model-ef2b2ab4d974