Summary of Lynch Lab Oct 25 2026 benchmark: Inception v3 125/200 vs GPT-6 Astra 84 and Claude Fable 5.1 85 on biopsy-proven lesions; discharge-threshold caveats—CNN released more melanomas at default threshold until matched to LLM sensitivity. n=200, single LLM run each.