1 article found
Tencent’s Hy4 Flash packs 224B parameters but activates only 12B per token. At $0.07/M input tokens, it’s rewriting inference economics for everyday AI work.