kog

Kog announces real-time LLM inference achieving 3000+ output tokens per second per request on standard datacenter GPUs, bringing high-speed inference previously limited to custom silicon to production hardware.