Google Gemma 4 leads local writeups

Gemma 4 looks like the safest default here because the thread values prose quality, instruction following, and clean formatting more than coding-agent behavior.

• –The comments consistently favor Gemma 4 31B for writing-heavy workflows and "local ChatGPT" style use.
• –Qwen3.6-27B is the strongest alternative: official model docs show a dense 27B model with 262k native context and multimodal support.
• –On 64GB unified memory, both are feasible at sensible quantization; the decision is mostly about output style and speed, not whether the model fits.
• –If your priority is markdown docs over tool use, start with Gemma 4 31B and keep Qwen3.6-27B as the backup when you want a more balanced generalist.