What GPU do you need for Qwen3.8, Qwen3.6, Gemma 4 and gpt-oss locally? (measured, Oct 2026)