Jeff Barg

We're looking into this.

Hey Ty - is this still ongoing? We bumped up a lot of our timeouts and have been closely monitoring any Claygent issues. We're fairly confident that things should be operating well, but I want to make sure we're not missing something that you're still seeing.

Does it show up as a timeout?

Yes taking a look now - Ty, do you have any error messaging / details on this? It'd be very helpful for us.

Also, as a side note — I see you're using Claude 3 Opus -- in general, I'd definitely recommend moving those workloads over to Sonnet 3.5, 3.7, or 4. Even Sonnet 3.5 should perform ~strictly better than Claude 3 Opus from what we've seen (& it's more cost effective), but I'll of course defer to what you've seen performs best for your use case.

That's what I'll try to understand -- In general, Anthropic models don't have the same structured outputs support than OpenAI does & they can output things that don't conform to the schema. So it's possible this is a model limitation. I think you can do a few things to improve this (i.e. in the examples in the prompt here, I think the example outputs have the extra s (i.e. they say opsbodyB instead of opbodyB). I'll keep working to confirm this, I want to rule out there being a problem from Clay's platform.

Gotcha - the issue is definitely the name mismatch, but I will look into why/if the model is outputting something inconsistent here

I think you may have renamed the fields slightly - the output field there is opsbodyA but the column is expecting opbodyA -- can we try re-running?

TAL, thanks for flagging

Just confirming the fix here made it out, and the table you linked now works with Anthropic models when retrying ✅