Skip to main content

Embedding Benchmark

Measure embedding latency, throughput, and dimensions across Voyage AI models.

Run​

vai benchmark embed

What It Measures​

  • Latency per embedding call (p50, p95, p99)
  • Throughput (tokens per second)
  • Output dimensions per model
  • Comparison across multiple models

Customize​

# Test specific models
vai benchmark embed --models voyage-4-large,voyage-4-lite

# More iterations for reliable results
vai benchmark embed --iterations 10

# JSON output
vai benchmark embed --json

Sample Output​

The benchmark uses built-in sample texts (10 documents about common tech topics) so you don't need any setup. Results show a formatted table comparing each model's performance.

Tips​

  • Run benchmarks multiple times — API latency varies with server load
  • Network latency dominates for single-text embeds; batch throughput is a better measure of model speed
  • Use --json to track performance over time