For a long time, the AI outbound calling industry has lacked a recognized "standard of measurement." Today, Agora has partnered with Meituan and xbench to officially launch an intelligent evaluation benchmark for AI outbound calling scenarios—VoiceAgentEval, aiming to fill the industry gap and accelerate the transformation of AI models toward real business scenarios.

Reject lab data, test real business "true gold."

The biggest highlight of VoiceAgentEval is its "practicality":

Wide coverage: It covers 30 sub-scenarios across six business areas, striving to recreate the most realistic market demands.

Real corpus: Built based on real outbound calling business data, it abandons traditional rigid scripts.

Two-dimensional evaluation: Not only does it check whether the logic of text generation is correct, but it also adds a voice dimension evaluation, comprehensively examining the AI's overall performance in conversations.

150 simulated dialogues, smoother AI practice.

To test the model's task adherence and general interaction capabilities, the evaluation framework built 150 virtual dialogue scenarios through a user simulator. This is like giving AI a series of "mock exams," evaluating whether it can steadily advance the business process when facing different user feedback.

Who are the current top performers in AI outbound calling?

According to the information, through the preliminary screening of this evaluation standard, the top three models in comprehensive performance in AI outbound calling scenarios have already been identified. This result not only sets a technical benchmark for the industry but also provides authoritative reference for related companies (such as Beijing San Kuai Technology Co., Ltd., etc.) in subsequent AI business deployment.