Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard
Points and comments are a snapshot, not live.
Article body wasn't reachable. The HN discussion summary is below.
Points and comments are a snapshot, not live.
Article body wasn't reachable. The HN discussion summary is below.
What commenters are saying
Many commenters treated the result as probable contamination or benchmark saturation, noting the underlying GVS5H method was posted a week prior. A split emerged: some argued the harness makes local models competitive; others warned that Bonsai 2-bit quants degrade quality and that swarm strategies cannot overcome that loss. Multiple users reported Bonsai 2-bit models are notably dumber in practice. A few pointed to Qwen-3.8-Flash-Next as a better candidate for local replication.