AutomationBench-AA - HR - Raw Objective Completion: leaderboard
Artificial Analysis's implementation of Zapier's AutomationBench restricted to its HR tasks: the share of task objectives an agent completes in simulated SaaS applications, counted before the guardrail-violation adjustment used by the AutomationBench-AA headline. Reported 0-100 (%); higher is better.
Metric: Raw Objective Completion (%). Source: artificialanalysis.ai. Saturation forecast: Around February 2027. 243 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Wonder Full (Max) | 83.32 | #12 (Muse Wonder Full) |
| 2 | Gemini 4 Argon (High) | 81.99 | #9 (Gemini 4 Argon) |
| 3 | Gemini 4 Carbon (High) | 81.81 | #1 (Gemini 4 Carbon) |
| 4 | Kimi K3.1 Pro | 79.44 | #17 |
| 5 | Muse Wonder (Max) | 78.45 | #18 (Muse Wonder) |
| 6 | MiniMax-M3.1-Flash | 77.5 | #52 |
| 7 | DeepSeek V4.1 Flash (1017) (High) | 77.41 | #27 (DeepSeek V4.1 Flash (1017)) |
| 8 | GPT-6 Astra (High) | 77.08 | #6 (GPT-6 Astra) |
| 9 | Grok 4.6 (High) | 76.94 | #29 (Grok 4.6) |
| 10 | GPT-6 Astra (Max) | 76.89 | #6 (GPT-6 Astra) |
| 11 | GPT-6 Astra (xHigh) | 76.37 | #6 (GPT-6 Astra) |
| 12 | GPT-6.1 Sol (High) | 76.37 | #7 (GPT-6.1 Sol) |
| 13 | Ling 3.1 Flash | 76.18 | #75 |
| 14 | GPT-6.1 Sol (xHigh) | 76.04 | #7 (GPT-6.1 Sol) |
| 15 | Qwen 3.8 Max (0902) | 75.76 | #36 |
No result here: #3 Claude Opus 5.5, #10 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=automationbench-aa-hr-raw-objective-completion · How It Works · Data refreshed daily, snapshot 2026-11-05.