LightSeek Foundation Releases TokenSpeed, an Open-Source LLM Inference Engine Targeting TensorRT-LLM-Level Performance for Agentic Workloads

LightSeek Foundation previewed TokenSpeed, an MIT-licensed open-source LLM inference engine for agentic coding workloads.

It uses compiler-assisted parallelism, a safety-focused scheduler, and modular kernels to improve long-context serving performance.

The engine is designed to boost throughput and responsiveness for multi-turn requests common in tools like Claude Code, Codex, and Cursor.

Better inference efficiency could lower latency and raise capacity for software development infrastructure.

TokenSpeed enters a competitive space alongside systems like vLLM and TensorRT-LLM, including on NVIDIA Blackwell GPUs.