# I moved our extraction step off Claude — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/d8fc63cfff49211b

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-09-30 15:33 ingested 2026-10-01 02:38

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1wu82wz/i_moved_our_extraction_step_off_claude/)

## retained text



```
For a bit more context for you guys we had one workflow where Claude was doing everything from reading the input to pulling out a handful of fields and then handling the parts that needed some reasoning.

I started looking at the requests separately and realized the extraction part was pretty boring with the same fields every time, structured output and easy to validate so I moved that step to a cheaper model and kept Claude for the part where the input gets messy or we need it to make a judgment.

It’s worked better than I expected but now I’ve created another problem for myself. There are some inputs where the cheaper model clearly isn’t going to do a good job and we end up trying it anyway then failing validation and sending the same thing to Claude afterward so those requests are doing more work than they did before.

I can already spot a few things that tend to cause it like longer documents and less consistent formatting but it’s not reliable enough that I want to hardcode a bunch of rules around it. Anyone that is splitting a workflow like this are you deciding which model gets the request upfront or letting the cheaper model try first and escalating when it fails?
```



document fetched 2026-10-01 00:24 · canonical https://reddit.com/r/ClaudeAI/comments/1wu82wz/i_moved_our_extraction_step_off_claude
