We’re running into two issues:
- As per user agent usage has surged, we’ve seen a very large increase in our slow pool load. The slow pool was conceived years ago when people wanted to make 200 requests per month, not thousands.
- As models have started to get more work done (tool calls, code written) per request, their cost per request has gone up; Sonnet 4 costs us ~2.5x more per request than Sonnet 3.5 (and writes more code / does more ambitious tasks!).
To fix each of these, we’re currently planning on instituting the following in a few days:
Sunsetting the slow poolEDIT: We’re going to rethink this.- Pricing Sonnet 4 at API cost converted to requests included in the plan (i.e. $0.04 API cost = 1 requests)
We wanted to give you all a heads up and solicit feedback once more. Open to other suggestions as well!