Hy3 (Tencent) is a 295.0B parameters MoE (21.0B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 167.45 GB VRAM for Q4 (weights 162.25 GB + KV 4.2 GB + 1 GB overhead) and 595.2 GB for FP16. About 3 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 242.4 tok/s (estimated); MI300X 192GB ≈ 160.6 tok/s (estimated); Mac Studio M2 Ultra 192GB ≈ 24.2 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.
| Hardware | VRAM | Quant | tok/s | Fits? | tok/s/$ |
|---|---|---|---|---|---|
| H200 141GB SXM5 | 141 | Q4 | 294.7 est. | No | 0.008 |
| B200 192GB SXM | 192 | Q4 | 242.4 est. | Yes | 0.006 |
| Gaudi 3 128GB | 128 | Q4 | 165.7 est. | No | 0.013 |
| MI300X 192GB | 192 | Q4 | 160.6 est. | Yes | 0.011 |
| H100 80GB SXM5 | 80 | Q4 | 66.2 est. | No | 0.002 |
| A100 80GB SXM4 | 80 | Q4 | 40.3 est. | No | 0.003 |
| MacBook Pro M4 Max 128GB | 128 | Q4 | 27.6 est. | No | 0.006 |
| Mac Studio M2 Ultra 192GB | 192 | Q4 | 24.2 est. | Yes | 0.003 |
| Mac Studio M3 Ultra 96GB | 96 | Q4 | 23.3 est. | No | 0.006 |
| MacBook Pro M3 Max 128GB | 128 | Q4 | 20.2 est. | No | 0.005 |
| B200 192GB SXM | 192 | FP16 | 19.8 est. | No | 0.0 |
| MI300X 192GB | 192 | FP16 | 13.1 est. | No | 0.001 |
| RTX 6000 Ada 48GB | 48 | Q4 | 6.8 est. | No | 0.001 |
| L40S 48GB | 48 | Q4 | 6.1 est. | No | 0.001 |
| RTX A6000 48GB | 48 | Q4 | 5.5 est. | No | 0.001 |
| Mac mini M4 Pro 64GB | 64 | Q4 | 3.5 est. | No | 0.002 |
| Mac Studio M2 Ultra 192GB | 192 | FP16 | 2.0 est. | No | 0.0 |