Alipay-PIBench: A Realistic Payment Integration Benchmark for Coding Agents

Alipay-PIBench evaluates coding agents on repository-level payment integration tasks across nine products and 18 scenarios. The benchmark uses deterministic static, unit, and integration checks alongside LLM-assisted semantic assessment. Testing six models shows that providing specific integration skills increases mean rubric pass rates by an average of 10.31 percentage points.