NousResearch Hermes adapter lifts Qwen3.8-27B non-thinking perf 14 points, cuts tokens 3.5-4.4x
yb2698 · x · 2026-11-19
NousResearch released a Hugging Face Model Adapter applying Hermes to Qwen3.8-27B, aimed at speeding up locally-run agents.
Key numbers: SWE-Bench Pro (hard 51) pass@1 jumps 29.4% → 43.1% (45.1% → 51.0% with xhigh thinking); Terminal-Bench 2.1 pass@1 rises 44.7% → 58.8% (slightly down to 57.6% with xhigh thinking). With thinking off, it matches the base model at xhigh thinking while using 3.5-4.4x fewer tokens and fewer tool calls — a meaningful efficiency win for local agent setups.
More from coding & agent
- Kong launches Volcano, a 'self-healing' cloud for agents after handling 500T AI tokens a year — EXM7777 · 2026-11-19
- Grok bot trained on Dan Koe's system turns one idea into a week of content — VibeMarketer_ · 2026-11-19
- Matt Pocock unveils VECTOR method for adopting AI on any task, seeks early feedback — mattpocockuk · 2026-11-19
- Meta's MLE agents ran 1,097 experiments on production rankers with 29.6% acceptance rate — dair_ai · 2026-11-19
- 11 system security concepts every engineer should know, from OAuth to Zero Trust — goyalshaliniuk · 2026-11-19
- Open-source Nango connects AI agents to 1,200+ APIs with one auth flow — aftahi_ai · 2026-11-19