AnalysisAI ModelsJune 19, 2026

Qwen3.6-27b-FP8 runs 262K context at 55 tok/s on $1800 GPU setup

A user runs Qwen/Qwen3.6-27b-FP8 with 262K context and BF16 KV cache on 4x 5060 Ti 16GB GPUs using P2P, achieving 55 tok/s for an $1800 GPU cost. The setup is described as single-user inference only.

1 source