ยง Infrastructure ยท Live Traffic Splitting

Traffic Splitting

Route live production traffic between model versions and compare performance in real-time

LIVEpolling every 4s
A: 90%/B: 10%
90%
๐Ÿ”ต
Model B is currently outperforming on quality + latency. Consider increasing Model B's traffic share.
Total Requests
Model A
10,824
Model B
1,343
B vs A90%
Error Rate
Model A
1.15%
Model B
0.56%
B vs A51.3%
Quality Score
Model A
95%
Model B
97.8%
B vs A2.9%
Tokens / sec
Model A
442
Model B
505
B vs A14.3%
Collecting live data...
๐Ÿ’ก How it works: Adjust the traffic slider and click "Apply Split" to persist the routing policy. Live metrics update every 4s. For statistical significance, collect at least 1,000 requests per variant before making deployment decisions.