I could see better TTS, streaming APIs, and context caching…
- Whisper turbo has already dropped as you wrote.
- I think browsing is unlikely. They don’t want to be responsible for developers’ web traffic.
- I don’t think a new knowledge update is big enough to warrant keeping it under wraps for DavDay.
- Cheaper prices would be great, but I can’t see it happening absent a new, more efficient model, and they’ve already said no new models, though they could be misleading.
- o1 and Sora would count as new models, so if we take OpenAI at their word, they’re not coming today.
My guess might be something to do with the reasoning models and maybe more control over the internal reasoning mechanism, like a dial for the amount of compute to spend on reasoning, or adding function calling/tool use, or remedying any of the other “beta limitations,”
