SOTA · inference tokens/sec/$ · LLaMA-3-70B, 8×H200, 2K ctx, BF16
$/hr from signed cloud price feed; tokens/sec measured on the sealed rerun cluster.
loading...
loading...
click to expand · verify re-checks sigs + inclusion
Systems · live
Whole stacks rather than parts: a model, a runtime, a driver and an accelerator, benchmarked together on sealed hardware. Component-level numbers do not compose, so these are measured end to end. Served live from .
SOTA · inference tokens/sec/$ · LLaMA-3-70B, 8×H200, 2K ctx, BF16
$/hr from signed cloud price feed; tokens/sec measured on the sealed rerun cluster.