GPT-4o ☆

GPT-4o merged modalities into a single end-to-end model — real-time voice conversation with ~300 ms latency, native image understanding and generation hooks — and became ChatGPT's free default in 2024, the moment multimodal stopped being a demo.