AutoBuddy-Bench - End-to-End: leaderboard
AutoBuddy-Bench v1, same 120-task simulated workplace, three runs per backbone, native End-to-End on 0–1 scale. Tasks discovered from workplace history, then executed through supplied tools. Execution is conditional on correctly discovered ground-truth tasks; precision/recall/F1 and end-to-end use their separately defined denominators. Source standard deviation retained in primary table, not treated as independent sampling SE. Exact source backbone and AutoBuddy harness remain a run identity; no model release is inferred.
Metric: End-to-End (0–1). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | claude-opus-5.5 (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.7 | |
| 2 | claude-opus-5 (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.68 | |
| 3 | gpt-5.6-sol (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.65 | |
| 4 | grok-4.7 (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.64 | |
| 5 | gpt-5.5 (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.63 | |
| 6 | DeepSeek-V4-Pro (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.59 | |
| 7 | gpt-6.1-sol (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.54 | |
| 8 | claude-opus-4.8 (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.52 | |
| 9 | gpt-5.4 (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.52 | |
| 10 | DeepSeek-V4.1-Flash (AutoBuddy Run: protocol=120-task-world; runs=3; harness=AutoBuddy) | 0.49 |
Interactive version: theaggregate.ai/benchmark?slug=autobuddy-bench-end-to-end · How It Works · Data refreshed daily, snapshot 2026-11-12.