Nous Research ships Hermes adapter for Qwen3.8-27B, big agentic benchmark gains
yb2698 · x · 2026-11-20
Nous Research announced a Hermes adapter for Alibaba's Qwen3.8-27B, claiming major agentic-coding gains:
- SWE-Bench Pro (hard 51) pass@1: 29.4% → 43.1%; with xhigh thinking: 45.1% → 51.0%
- Terminal-Bench 2.1 pass@1: 44.7% → 58.8% (57.6% with xhigh thinking)
- With thinking off, it matches the base model at xhigh while using 3.5–4.4x fewer tokens
Teknium amplified the release with a "give it a try" nudge. The adapter is available on GitHub/HF.
More from coding & agent
- Dev: Antigravity runs 10 hours with no status hint, laggy side chat hides crashes — brianryhuang · 2026-11-20
- emulate ships local OpenAI & Anthropic API emulator for deterministic zero-cost AI testing — cramforce · 2026-11-20
- Dev finds wiring new Google models into n8n painful, sticks with OpenRouter — ColleenMBrady · 2026-11-20
- Dev observes AI models now auto-apply GI, volumetrics and AA without being asked — msfeldstein · 2026-11-20
- BMW engineers assign trust per step in agent workflows, with code-enforced permission checks — shoumikgoswami · 2026-11-20
- Designer benchmarks every new Gemini by having it rewrite his site's hero shader — AIandDesign · 2026-11-20