AgentRewardBench Leaderboard

💾Code	📄Paper	🌐Website
🤗Dataset	💻Demo	🏆Leaderboard

This is the leaderboard for the AgentRewardBench. The scores are based on the results of the agents on the benchmark. We report the precision score. Open an issue to submit your results to the leadeboard. We will review your results and add them to the leaderboard.

Judge	Overall	AB	VWA	WA	Work	Work++	URL	Author
Rule-based	83.8	25	85.2	79	100	83.3	https://arxiv.org/abs/2504.08942	Lù et al.
WebJudge	73.7	66.7	69.8	72.6	92.3	75	https://arxiv.org/pdf/2504.01382	Xue et al.
AER-C (GPT-4o)	67.7	83.3	56	68.8	100	66.7	https://arxiv.org/abs/2504.08942	Lù et al.
AER-V (GPT-4o)	67.6	83.3	61.2	67.6	96.4	59.3	https://arxiv.org/abs/2504.08942	Lù et al.
NNetNav (Llama-3.3 70B)	52.5	20.8	54.5	54.3	77.3	43.2	https://arxiv.org/abs/2504.08942	Lù et al.
Claude 3.7 S. (A)	68.8	87.5	61	69.3	85	66.7	https://arxiv.org/abs/2504.08942	Lù et al.
GPT-4o (A)	69.8	77.8	63	70.2	94.6	63	https://arxiv.org/abs/2504.08942	Lù et al.
GPT-4o Mini (A)	61.5	80	57.9	63.5	84.2	49.4	https://arxiv.org/abs/2504.08942	Lù et al.
Llama 3.3 (A)	67.7	75	59.6	68.2	94.3	62.7	https://arxiv.org/abs/2504.08942	Lù et al.
Qwen2.5-VL (A)	64.3	72.7	59.3	63.6	87.2	60.3	https://arxiv.org/abs/2504.08942	Lù et al.
Claude 3.7 S. (S)	69.4	71.4	64.8	69.3	85.3	66.7	https://arxiv.org/abs/2504.08942	Lù et al.
GPT-4o (S)	68.1	77.8	60.7	69.9	93.8	59.6	https://arxiv.org/abs/2504.08942	Lù et al.
GPT-4o Mini (S)	64.5	80	57.4	66.9	90.3	54.8	https://arxiv.org/abs/2504.08942	Lù et al.
Qwen2.5-VL (S)	64.5	70	58.5	62.9	93.8	64.4	https://arxiv.org/abs/2504.08942	Lù et al.