Post

A Jev that can ask Opus for help gives Opus accuracy at 4x lower cost

Hightouch tested routing uncertain identity-resolution judgments from a cheap decision model to a larger model. On its easier test, Jev plus Opus achieved 98.8% accuracy at $0.88 per 1,000 identities versus Opus alone at 99.0% and $3.86.

On the harder set the hybrid cost $2.94 versus $5.39 for Opus, with comparable reported accuracy. The authors attribute the result to confidence calibration; it is a task-specific experiment, not a general guarantee. HN readers discuss applying similar routing to tool checks and context management.