What Do You Need to Run Hunyuan Locally? VRAM by Version
Hunyuan currently has 2 versions you can deploy locally. The smallest, Hy3, needs 10GB of VRAM at Q4; the largest, Hy4 preview, needs 18GB. The cheapest device that gets one running is the Tesla V100 16GB, at about $289 used.
VRAM needed for each Hunyuan version
2 versions| Version | Released | Context | |||
|---|---|---|---|---|---|
| Hy3MoE | 295B (21B active) | Jul 6, 2026 | 256K | 10plus 160 GB of RAM | Tesla V100 16GB$289 · 16 GB · Experts offloaded |
| Hy4 previewMoE | 770B (49B active) | Aug 28, 2026 | 1M | 18plus 415 GB of RAM | Radeon RX 7900 XT$690 · 20 GB · Experts offloaded |
Minimum VRAM = Q4_K_M weights + KV cache for an 8K context + 1GB runtime overhead, rounded up. The cheapest device is the first one that fits when sorting by median used price on the current price basis.
About Hunyuan
Hunyuan is published by Tencent under Apache-2.0. This page covers 2 versions ranging from 295B to 770B parameters. New official releases are scanned weekly; architecture parameters and sourced or estimated weight sizes are reviewed before publication.
VendorTencent
LicenseApache-2.0
Official sitehttps://hy.tencent.ai/
Hugging Face orgtencent