Streaming in the context of OpenAI APIs refers to the model's answer being transmitted by chunks as it's being created, rather than waiting for the full answer.

https://platform.openai.com/docs/api-reference/streaming

Right now, the (Assistants API) announced at DevDay is publicly available. The problem is, it's practically commercially inviable to use it, as long answers could take up to 20 seconds to be crafted.

It's been a very requested feature on the OpenAI Developer Forum, and listed first on the to-do list for the upcoming API features https://platform.openai.com/docs/assistants/how-it-works/limitations

Timing on streaming is very relevant to our startup, and I guess it should be the same for a ton more.