Are Harnesses Worth the Cost?
Naresh Jain compared coding agents building the same API service with and without a behavioral harness. A frontier model, GPT-6 Sol, passed every external evaluation either way. A cost-optimized model, GPT-6 Luna, succeeded in only 3 of 10 unharnessed runs and all 10 with the harness. Harnessed Luna runs matched Sol's 10-for-10 reliability while costing 33% less per run ($0.273 vs $0.406).
It suggests a behavioral harness may let teams get reliable coding-agent results from smaller, cost-optimized models rather than frontier ones.


