Halyard
Logistics
71% lower extraction cost
Cut from extraction cost
Halyard reads 40,000 bills of lading a week. Routing cut the bill by two thirds without touching accuracy.
The problem
Every document went to one frontier model because that was the model that worked during the prototype. Cost scaled linearly with volume, and volume was growing 9% a month. The invoice hit $71,000 in June.
What they did
They wrote a 340-case eval suite from real documents, then set a routing policy: cheapest model that clears 0.94. Nothing else changed in the application.
What changed
atlas-8b now handles 78% of documents, with atlas-70b taking the scanned and handwritten ones. The invoice went to $20,600. Accuracy went up, from 94.1 to 94.3.
We were paying three vendors to disagree with each other. Now the router picks, the evals check, and the trace tells us why.
Priya Anand
Staff Engineer, Halyard
By the numbers
$71,000 to $20,600 monthly
78% of volume on the cheap model
94.1 to 94.3 accuracy
1M tokens a month, free forever.
No card, no trial clock, no sales call. Move to a paid plan when the invoice makes sense, or don’t.