"Fast inference for a wide range of models. Widely considered to be the best inference provider for Llama 3.1 8B, 70B, and 405B."