NousResearch's Hermes adapter lifts Qwen3.8-27B from 29.4% to 43.1% on SWE-Bench Pro
ariG23498 · x · 2026-11-19
NousResearch released a Hermes adapter for Alibaba's Qwen3.8-27B claiming large agentic-coding gains:
- SWE-Bench Pro (hard 51) pass@1: 29.4% → 43.1%; 45.1% → 51.0% with xhigh thinking
- Terminal-Bench 2.1 pass@1: 44.7% → 58.8%
- With thinking off, it matches the base model's xhigh-thinking performance at higher efficiency
Framed as a win for open science.
More from coding & agent
- NDepend open-sources a C# MCP Server with a complete tutorial and 35-minute video — adnan_hashmi · 2026-11-19
- Creator Studio ships with in-app assistants, MCP support, and open Keynote format — jasonwuishere · 2026-11-19
- Vercel's agent-desktop lets Claude Code drive native Mac apps via the accessibility tree — Arindam_1729 · 2026-11-19
- Anthropic publishes free 33-page guide on turning repetitive prompts into reusable Claude Skills — aftahi_ai · 2026-11-19
- Dev open-sources finance MCP server with OAuth 2.1, PKCE and plan-based tool gating — Noahowshhh · 2026-11-19
- n8n workflow evaluates LLMs with two cross-provider judges to avoid self-preference — ZtasnimS75 · 2026-11-19