1 article tagged with this topic
Qwen 2.4T hit 350 tokens/sec on NVIDIA's GB300 rack. Open-source is crushing inference costs — a 2026 China cloud price war looks likely.