Within 10k tokens, Hermes adapter solves 51% of TB-2.1 vs base's 28%
yb2698 · x · 2026-11-19
Within a 10k-token per-task budget, the Hermes adapter (thinking off) solves 51% of Terminal-Bench 2.1 and 39% of SWE-Bench Pro, while the base model with xhigh thinking manages only 28% and 0%. On tasks both solve, median tool calls drop from 23 to 7 on TB-2.1 and from 63 to 28 on SWE-Bench Pro.
More from coding & agent
- Ex-DeepSeek core member open-sources Linger, a personal AI agent that runs local models on your PC — FellMentKE · 2026-11-19
- Apple opens iWork to external LLMs: Pages, Numbers, Keynote gain MCP support — dsp_ · 2026-11-19
- Hermes adapter gets GGUF release: matches xhigh thinking at a quarter of the tokens — yb2698 · 2026-11-19
- Gym's chatbot invented a free membership-pause policy; team shares its guardrail lessons — Fit_Satisfaction541 · 2026-11-19
- Anthropic engineer shares free 30-minute Claude Code SDK + GitHub Actions automation walkthrough — aftahi_ai · 2026-11-19
- Snowflake Labs open-sources Cortex AI 101 notebook calling OpenAI models via SQL — adnan_hashmi · 2026-11-19