Tencent · 295B (21B active) · 専家混合（MoE）
Production-focused agentic MoE with strong coding, tool use and long-context reasoning 看看你的 GPU 或 Mac 跑不跑得動 Hy3——最低 164.8 GB，建議 274.7 GB。
2026-07256K context
専家混合（MoE）
専家総数： 192
启用専家： 8
启用参数： 21.0B
| 量化 | 位元 | VRAM | 品質 | 状態 |
|---|---|---|---|---|
| Q2_K | 2 | 94.9 GB | low | — |
| Q3_K_M | 3 | 132.7 GB | moderate | — |
| Q4_K_M | 4 | 151.6 GB | good | — |
| Q5_K_M | 5 | 189.4 GB | good | — |
| Q6_K | 6 | 227.2 GB | excellent | — |
| Q8_0 | 8 | 302.7 GB | excellent | — |
| F16 | 16 | 604.9 GB | lossless | — |
Can I run Hy3 locally?
- Can I run Hy3 locally?
- Hy3 needs about 164.8 GB of memory at a minimum and 274.7 GB recommended. Open this page to grade it against your GPU or Mac, then run it with runai, Ollama or LM Studio.
- How much VRAM does Hy3 need?
- At Q4_K_M, Hy3 uses about 151.6 GB of VRAM. Higher quants need more memory; lower quants fit tighter cards with a quality tradeoff.