AgentRewardBench (Accessibility Tree) - WorkArena++: leaderboard
Metric: Precision (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude 3.7 Sonnet | 66.7 |
| 2 | GPT-4o (2024-11-20) | 63 |
| 3 | Llama 3.3 70B Instruct | 62.7 |
| 4 | Qwen 2.5 VL 72B Instruct | 60.3 |
| 5 | GPT-4o Mini (2024-07-18) | 49.4 |
Interactive version: theaggregate.ai/benchmark?slug=agentrewardbench-accessibility-tree-workarena-plus-plus · How It Works · Data refreshed daily, snapshot 2026-09-26.