The Assistant API has been in beta for almost a year. The new structured response feature is absolutely critical for anyone building an application. Yet, it seems like the entire Assistant ecosystem is in disarray. For example, the rate limits are undocumented. After going back and forth with a support bot for weeks, a human finally responded with these specs -
GET: 1000/min
POST: 300/min
DELETE: 300/min
POST: “/v1/threads/_thread_id/runs”: 200/min
POST: “/v1/threads/runs”: 200/min
Can someone from OpenAI give us some insight into the Assistant roadmap for those of us building a platform with which we mean to serve clients?
So, it is almost 2 months since your post, and no response. Anyone got insight, or decently speculation, on why OpenAI is silent on this … and why such an extended Beta/(preview) period?
I’ve abandoned the Assistant API for most things and have gone back to the Chat Completion API.
The release notes for the Python SDK read like it’s one person who’s been tasked with making small, incremental improvements around the edges.
Someone needs to tell the engineering leadership that new models and features are cool, but you can’t run a company on cool - you run a company on reliable.
@OpenAl - from an outsiders perspective, your engineering appears to be in total disarray. Stop treating your tooling like a science fair project and focus on building a reliable toolset that companies can depend on.
Well, I wouldn’t go that far, but I would agree there are some issues to work out.
Personally, I love the Assistants API—it’s made it possible for me to pull off some serious magic with my app and in my domain. Technically, it requires more effort but the results are so worth it.
My hope is that they don’t abandon it all together. You can see more effort being put into other solutions (Chat, Swarm, Realtime), so maybe we’ll see some improvements this month.
It’s not a solution if it is in Beta. Both Swarm and Realtime are in beta and I suspect will stay in that state for a very long time if not forever.
OpenAI has become quite proficient at putting half baked software in beta while letting their production ready solutions rot.
There is a reason the world is complaining that there aren’t any meaningful AI solutions out there - its because companies like mine looking to build production ready solutions have to rely on public beta that move like molasses in January because their engineering teams continually get distracted by shiny objects.