Can I run Hy3 locally?
Hy3 by tencent has 299B parameters (21B active per token, 192 experts, 8 used per token). At 4-bit with an 8K context it needs about 173 GB of GPU memory. It fits on 6 of 65 GPUs and Macs we track.
Memory needed by quant and context length
GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from bartowski/Hy3-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.
| Quant | Quality | 4K ctx | 8K ctx | 32K ctx | 128K ctx |
|---|---|---|---|---|---|
| 2-bit (Q2_K) • | rough | 101 GB | 102 GB | 110 GB | 140 GB |
| 3-bit (Q3_K_M) • | usable | 130 GB | 131 GB | 139 GB | 169 GB |
| 4-bit (Q4_K_M) • | good, most popular | 172 GB | 173 GB | 180 GB | 210 GB |
| 5-bit (Q5_K_M) • | very good | 200 GB | 201 GB | 209 GB | 239 GB |
| 6-bit (Q6_K) • | near lossless | 241 GB | 243 GB | 250 GB | 280 GB |
| 8-bit (Q8_0) • | effectively lossless | 298 GB | 299 GB | 306 GB | 336 GB |
| 16-bit (F16/BF16) | original weights | 558 GB | 560 GB | 567 GB | 597 GB |
Which GPUs and Macs run Hy3?
Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).
Prefer an API?
Tencent: Hy3 is on OpenRouter at $0.132 per 1M input tokens and $0.528 per 1M output tokens. Price details and history →
Questions
How much VRAM does Hy3 need?
About 173 GB at 4-bit (Q4_K_M) with an 8K context: 170 GB of weights (real GGUF file size), 2.5 GB of KV cache and 0.6 GB of runtime overhead. At 8-bit it needs about 299 GB.
What is the smallest GPU that runs Hy3?
No single consumer GPU in our table fits it at Q4_K_M with 8K context; the smallest option that does is the Mac M3 Ultra 256GB (256 GB).
Can I run Hy3 on a Mac?
Yes. At Q4_K_M with 8K context it fits on a Mac M3 Ultra 256GB or larger, using the default macOS GPU memory limit.
Model facts
| Hugging Face repo | tencent/Hy3 |
| Parameters | 298,786,155,776 (299B) · ~21B active (computed from config) |
| Layers / hidden size | 80 / 4096 |
| Attention | GQA; 64 heads, 8 KV heads, head dim 128 |
| KV cache per token (FP16) | 320.0 KiB |
| Max context | 262,144 tokens |
| License | apache-2.0 |
| Released on HF | Jul 2, 2026 |
| Downloads (30d) / likes | 987,079 / 993 |
Sources: Hugging Face model API · config.json (tencent/Hy3) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.