AutomationBench-AA 1.0.6
Agents
Artificial Analysis' AutomationBench 1.0.6 series, identified by automation_bench_1_0_6 in its canonical data schema. The 657-task implementation reports partial objective completion with a zero score for guardrail violations, distinct from earlier source revisions.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Carbon | 77.8 |
| Gemini 4 Argon | 77.5 |
| Kimi K3.1 Pro | 73.3 |
| MiniMax M3.1 Flash | 72.5 |
| Claude Sonnet 5.5 | 71.8 |
| Muse Wonder Full | 71.3 |
| DeepSeek V4.1 Flash | 68.9 |
| GPT-6 Astra | 68.5 |
| Muse Wonder | 68.1 |
| Ling 3.1 Flash 1019 | 67.5 |
| Grok 4.6 | 67.0 |
| DeepSeek V4.1 Flash 1017 | 66.6 |
| GPT-6.1 Sol | 66.6 |
| Grok 4.7 | 65.6 |
| MiMo-V2.6-Flash | 64.1 |
| GLM-5.3 | 62.2 |
| Gemini 3.7 Flash | 62.0 |
| Ling 3.1 Flash | 61.7 |
| GPT-6 Sol | 61.7 |
| Gemini 3.8 Flash | 60.9 |
| GLM-5.3 Flash | 60.4 |
| GPT-6.1 Terra | 60.2 |
| GPT-5.6 Sol | 60.1 |
| Mistral Large 4 Preview | 59.9 |
| GPT-5.6 Terra | 59.6 |
| MiMo-V2.6-Pro | 58.6 |
| Kimi K3 | 58.3 |
| Grok 4.5 | 57.9 |
| Muse Spark 1.3 | 57.9 |
| Qwen3.8 2.4T A95B | 57.2 |
| DeepSeek V4 Pro 0813 | 56.7 |
| Claude Opus 5 | 56.6 |
| DeepSeek-V4-Pro | 56.3 |
| Qwen3.8 Max (0902) | 56.2 |
| Qwen3.8-Flash-Next | 55.9 |
| DeepSeek-V4-Flash-0731 | 54.0 |
| GPT-6 Luna | 53.2 |
| Gemini 3.6 Flash | 53.0 |
| Step 5 Preview | 51.0 |
| GPT-5.6 Luna | 50.2 |