OpenAI Assistants API Version 2 cliff notes:

Apr 18 (edited) in General discussion

There's a new tool called file_search that can handle 10,000 context and data files per assistant, which is considerably more than before. Plus, it's faster and smarter with searches.
Vector store objects are a new add. You drop a file in, and it gets set up for quick searches. Helps keep things organized and cuts down on embedding and vector DB hassle.
You've got more control over how many tokens the system uses each time, so you can keep an eye on costs.
There's a tool_choice option now. It lets you pick exactly which tool you want to use for each task.
You can make messages that look like they're from the assistant itself—handy for setting up conversations the way you want.
The setup for assistants and runs has more options, like tweaking how creative or precise the responses are.
Only the newer, custom-tuned versions of the GPT models are available to use.
The whole system supports streaming now, so it's smoother and more like real-time.
They’ve added some tools in their software kits to help manage streaming and checking for updates.

All in all, I think this is a great step in the right direction.

0 comments

skool.com/ai-developers-2734

Master generative AI to write code and build powerful, real-world solutions faster, with guidance in a community of like-minded builders.

Leaderboard (30-day)