What it is o1 preview is one of two o1 models in ChatGPT's model selector — the little smarter, a little slower option with more knowledge of the world. Its counterpart, o1 mini, is faster but knows less about the world and is better tuned for math and coding problems [^4ef67c54-9505-5bc4-a565-9579a83f2827]. Why it matters o1's breakthrough is Chain of Thought reasoning done natively: the model shows a step-by-step thought process before answering, which improves accuracy and reduces hallucinations [^4ef67c54-9505-5bc4-a565-9579a83f2827]. How this channel uses it In the Hermes build, models are chosen by specialty — and o1 preview gets the interpretation job: it analyzes the transcription of a photographed book page because its chain-of-thought reflection makes it better at interpreting text [^20a397df-10fb-596f-826a-f54787502261@234]. In that same build, GPT-4 handles image-to-text and o1 mini writes the code. When a model doesn't know about o1 itself, the fix is to feed o1 its own documentation so it understands how to call itself [^20a397df-10fb-596f-826a-f54787502261@302]. When to pick it - Pick o1 preview for questions that require knowing things about what's going on in the world — and for interpreting text [^20a397df-10fb-596f-826a-f54787502261@234].- Pick o1 mini when speed matters most, or for math and coding problems [^20a397df-10fb-596f-826a-f54787502261@268]. - o1 preview is the slower, smarter, more world-knowledgeable o1 model.- Its native Chain of Thought reasoning improves accuracy and reduces hallucinations.- In practice it is used for interpretation tasks, while o1 mini handles code and GPT-4 handles OCR.- Feeding models their own current documentation fixes blind spots.