Benchmark runs comparing cost and quality across models.

Tok/s vs cost benchmarks for running open models yourself, across GPUs and inference engines.