Hermes Index
Indexes · 2026-10-06
Nous Research's index of agentic capability inside its Hermes Agent harness: the unweighted mean of four pass@1 suite scores, its own Hermes Bench plus Nous's runs of TerminalBench 4 without its four GPU tasks, TerminalBench Science and SkillsBench with tasks whose public solutions a run found removed. Every model runs at reasoning effort high where it offers one. The suite runs are Nous's own and are not the maintainers' official results.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6.1 Sol | 57.1 |
| GPT-6 Astra | 56.3 |
| GPT-6 Sol | 44.1 |
| Grok 4.7 | 39.3 |
| GLM-5.3 | 39.2 |
| Kimi K3 | 39.2 |
| Gemini 3.8 Flash | 38.1 |
| Claude Haiku 5.5 | 37.0 |
| GLM-5.3 Flash | 34.9 |
| GPT-6 Luna | 33.9 |
| Step 5 Preview | 32.8 |
| Hy4 Preview | 26.1 |
| Ling 3.0 Flash | 21.6 |