InferenceX
An open-source, continuously updated benchmark research platform that tracks real-world inference performance of popular open-source serving frameworks such as vLLM, SGLang and TensorRT-LLM across current GPU hardware and models.
InferenceX, formerly InferenceMAX, is an inference performance research platform that continually analyzes and benchmarks popular open-source inference frameworks used by major AI operators, tracking real performance in near real time as those software stacks improve. Only the maintainer's own repository holds the official InferenceX result; forks may differ in benchmark setup and machine quality and must be explicitly labeled unofficial.
It publishes continuous benchmark dashboards for current models, including Kimi K3, MiniMax M3, DeepSeek V4 and GLM5, comparing hardware such as NVIDIA's GB200 NVL72, GB300 NVL72 and B200 against AMD's MI355X, with support for further accelerators planned. Recent additions include an agentic benchmark suite covering realistic million-plus-token, multi-turn context, and native multi-token prediction dashboards for newer model releases.
It is open source under the Apache 2.0 license and written in Python, with a free public dashboard.