How GPT-4 appears in this material In the captured material, GPT-4 is used for one concrete job: converting an image into text. In the Hermes build — an SMS reading companion — a photo of a book page is texted in, and GPT-4 handles the image-to-text step because it is multimodal and strong at OCR[^9Zmhe6_T-xU@211]. One model per specialty GPT-4 is deliberately paired with other models, each chosen for its strength: - GPT-4 → image-to-text (OCR)[^9Zmhe6_T-xU@211] - o1 preview → interpreting the transcription, since its chain-of-thought reflection makes it better at reading text[^9Zmhe6_T-xU@234] - o1 mini → writing the code: smaller and faster, better at programming, though it knows less about the world[^9Zmhe6_T-xU@268] The takeaway is architectural: pick the right model for each job rather than forcing one model to do everything. - GPT-4 is used as the image-to-text model in the Hermes app because it is multimodal and strong at OCR.- It sits alongside o1 preview (interpretation) and o1 mini (coding), illustrating a multi-model architecture where each model handles its specialty.