Tencent Hy — local AI model by Tencent
Written by Jakub Rusinowski · Last updated
Tencent's open-weight flagship language models (the line formerly called Hunyuan): Hy3, a 295B mixture-of-experts with 21B active per token, and the 770B Hy4 preview. Both are Apache-2.0 and carry a built-in multi-token-prediction layer for speculative decoding. For Tencent's translation models, see Hy-MT2.
Variants
The smallest Tencent Hy variant needs about 181 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| Tencent Hy3 295B-A21B → 295B (21B active) | ~181.2 GB | 262,144 | ollama run tencent-hy | |
| Tencent Hy4 Preview 770B-A49B → 770B (49B active) | ~471.7 GB | 1,048,576 | ollama run tencent-hy |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run Tencent Hy locally
Install Ollama, then pull the tag.
ollama run tencent-hyPick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: Tencent Hy3 295B-A21B, Tencent Hy4 Preview 770B-A49BRecommended GPU
The cheapest catalogued GPU that runs Tencent Hy locally (min 181 GB VRAM) is the Apple M3 Ultra (512 GB).
Tencent Hy — frequently asked questions
How much VRAM does Tencent Hy need?
Tencent Hy needs about 181 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Tencent Hy3 295B-A21B (181 GB, Q4_K_M); Tencent Hy4 Preview 770B-A49B (472 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Tencent Hy on an RTX 4090 (24 GB)?
Tencent Hy's smallest variant needs about 181 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
What quantization should I use for Tencent Hy?
Q4_K_M is the best balance of quality and VRAM for Tencent Hy in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Tencent Hy with Ollama?
Tencent Hy has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.