Ran a classic(medival europe) fantasy RP/agentic benchmark across 8 local models Qwen3.6-27B held up better than its size suggests

Threw together a benchmark suite (quest completion, scene endings, item/time tracking, character detection, storytelling, drafting) and ran it across 8 models people talk about a lot on here. Judged with an external LLM grader, N varies per category (shown on the chart). Overall pass rates: ge

中文全文翻译正在补齐，可先查看下方英文原文。

Threw together a benchmark suite (quest completion, scene endings, item/time tracking, character detection, storytelling, drafting) and ran it across 8 models people talk about a lot on here. Judged with an external LLM grader, N varies per category (shown on the chart). Overall pass rates: ge