Benchmarks / Agentic
Hermes Bench
Uses mean reward across 150 Hermes Bench tasks in the Hermes Agent harness. Score: mean task reward. Collected by Jake Boggs' benchmark dashboard.
Hermes Bench
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Opus 5.5 | max | 76.7% |
| 2 | Grok 4.7 | max | 74.4% |
| 3 | GPT-6 Astra | max | 73.7% |
| 4 | GPT-6 Sol | max | 73.5% |
| 5 | Sonnet 5.5 | max | 73.5% |
| 6 | GLM 5.3 | max | 71.8% |
| 7 | Kimi K3 | max | 71.5% |
| 8 | DeepSeek V4.1 Flash | max | 70.7% |
| 9 | Qwen 3.8 Max | max | 70.5% |
| 10 | GPT-6 Luna | max | 70.5% |
| 11 | GLM 5.3 Flash | max | 68.9% |
| 12 | Gemini 3.8 Flash | max | 62.2% |
| 13 | Hy4 | – | 58.6% |
| 14 | Ling 3.0 Flash | – | 53.8% |