A local model finds the bugs, Opus throws out the noise
AISLE's nano-analyzer benchmark rerun on our Mac Studio with Qwen 3.6 35B A3B, then Claude Opus 4.6 as a paid second opinion: detection, false alarms, time and cost.
We test EpicScan against real, published vulnerabilities before we scan anyone's code, and we publish what we measure, including the misses.
AISLE's nano-analyzer benchmark rerun on our Mac Studio with Qwen 3.6 35B A3B, then Claude Opus 4.6 as a paid second opinion: detection, false alarms, time and cost.