Five small models against a frontier model: the team-solve result — Memdoor docs

Five Qwen3.8-27B agents, run as a Memdoor workflow, scored a mean of 9.8 of 10 on the first ten problems of GVS5H's LiveCodeBench Hard list over five passes.

Memdoor · Docs · Enterprise