EnterpriseOps-Gym-AA
Agents · 2026-07-08
Artificial Analysis' independent implementation of ServiceNow's EnterpriseOps-Gym evaluates LLM agents on 1,117 distinct oracle-mode tasks across eight business domains, with three runs per task (3,351 trajectories). The headline score averages final-state task success across those repeated runs.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 51.1 |
| Gemini 3.5 Flash | 50.1 |
| Claude Opus 5 | 47.5 |
| Muse Spark 1.1 | 47.2 |
| GPT-5.5 | 46.6 |
| Kimi K3 | 45.3 |
| Qwen3.7-Max | 45.0 |
| Sonnet 5 | 44.7 |
| Opus 4.8 | 44.0 |
| GPT-5.6 Sol | 42.9 |
| GLM-5.2 | 42.7 |
| Gemini 3.1 Pro Preview | 42.2 |
| Grok 4.5 | 40.8 |
| Qwen3.7-Plus | 40.6 |
| DeepSeek-V4-Pro | 40.4 |
| Kimi K2.7 Code | 40.2 |
| DeepSeek-V4-Flash | 39.6 |
| Kimi K2.6 | 38.5 |
| Inkling | 38.1 |
| GPT-5.4 Mini | 34.6 |
| Mistral Medium 3.5 | 33.7 |
| MiniMax M3 | 32.1 |
| GPT-5.4 Nano | 31.5 |
| Qwen3.5 397B A17B | 30.9 |
| Haiku 4.5 | 29.4 |
| Step 3.7 Flash | 29.1 |
| Nemotron 3 Ultra 550B A55B | 28.9 |
| Gemma 4 31B IT | 28.3 |
| Gemini 3.1 Flash-Lite Preview | 28.0 |
| Grok 4.3 | 26.3 |
| gpt-oss-120b | 25.5 |