GPU inference
performance
benchmarks

Methodology note: Throughput figures (tokens/sec) are measured under controlled load conditions and may vary in production based on concurrent requests, prompt complexity, and provider infrastructure. Pricing data reflects list rates and may not include enterprise discounts. Rows marked ✱ indicate internally generated or synthetic benchmark data used to fill gaps where third-party measurements are unavailable — these should be treated as estimates. Context degradation curves are modelled from available data points. GPU specs sourced from vendor datasheets. For full methodology see seeth.gudetee@gmail.com.