CoralBricks
モデルアグリゲーターCoralBricks serves open models (GLM, DeepSeek) quantized to NVFP4 / MXFP4 on its own GPUs behind an OpenAI-compatible API (Chat Completions, Responses and background mode), billed per token in USD from prepaid credits. Cached input is billed at $0 on every model; novel input under prompt-cache retention bills at a separate cache-write rate.