The post releases two GGUF quantized models—Tencent's Hy3 (295B MoE with 21B active parameters, 262K context) and NVIDIA's Nemotron‑Labs‑Audex‑30B‑A3B (audio‑capable