Floor for local meeting summarization on a 6GB GPU: Qwen3.5 0.8B works in 57s, Granite 4 350M hallucinates

Floor for local meeting summarization on a 6GB GPU: qwen3.5:0.8b works at 57s, Granite 4 350M hallucinates

The author tested VoiceFlow 1.6.0 on an RTX 3060 Laptop 6GB, where Qwen3.5 0.8B summarized a 4-minute meeting in 57 seconds with 16K context, while Granite 4 350M returned summaries in 0.6-2.8 seconds but fabricated Binance and Star Trek content.

Why it matters: HKR-H/K/R all pass: the hook is concrete, the test reports hardware/context/timing, and local meeting summarization hits privacy and cost nerves. Single Reddit experiment limits authority, so 73 featured.