Performance

Real-world benchmarks for the Chinese AI ecosystem.

We don't just route tokens; we measure them. Agent Score is our proprietary benchmarking system that tracks the performance, latency, and reasoning quality of domestic LLMs in real-time.

? Real-Time Metrics (Live Lab)

• Latency Index: Track the TTFT (Time-To-First-Token) across DeepSeek, Kimi, and Qwen via our edge nodes.
• Reasoning Accuracy: Monthly evaluation of model performance on complex logic and coding tasks.
• Provider Stability: A 30-day rolling uptime report for every upstream provider in our network.

---

? Why we do this? In a rapidly evolving AI market, performance is a moving target. We test 24/7 so you can focus on building, knowing your app is always powered by the optimal model.