Hunyuan-Large.
Tencent's open-weight MoE — 389B total, 52B active. Largest open MoE at launch.
1× AMD MI325.
Most-aggressive quantisation we have a working recommendation for. Lower precision = less VRAM = cheaper hardware, at a small accuracy cost.
Cheapest hosted endpoints.
| Provider | Access | $/M in | $/M out | |
|---|---|---|---|---|
| Self-hosted on rented GPU cluster | self hosted | — | — | Run yourself → |
| Self-hosted on rented GPU cluster | self hosted | — | — | Run yourself → |
| Tencent Cloud | api direct | — | — | Launch ↗ |
| Tencent Cloud | api direct | — | — | Launch ↗ |
Frequently asked.
How do I run Hunyuan-Large?
Where can I access Hunyuan-Large?
How much does it cost to run Hunyuan-Large?
Is Hunyuan-Large open-source or proprietary?
Cheapest hardware per quantisation.
Each row is one quantisation tier (the same weights compressed differently). Lower precision → lower VRAM → cheaper hardware, at the cost of small accuracy loss. $/hr refreshed hourly from each provider's API.
| Quantisation | Cheapest GPU config | Total VRAM | Live $/hr | tokens/sec | |
|---|---|---|---|---|---|
|
FP16
FP16 — half precision (default)
|
1024 GB | — | — | Compare → | |
|
FP8
FP8 — 8-bit float (Hopper / Blackwell)
|
512 GB | — | — | Compare → | |
|
INT4
INT4 — 4-bit integer (~4× VRAM saving)
|
256 GB | — | — | Compare → |
About Hunyuan-Large.
Hunyuan-Large was the largest open-weight Mixture-of-Experts model when Tencent released it in November 2024 — 389B total parameters with 52B active per token. Trained on 7T+ tokens with a strong English/Chinese balance. The model card highlights long-context (256K) and high tool-use accuracy. License is permissive for commercial use under Tencent's Hunyuan Community terms (revenue threshold above which a commercial agreement is required). Used as a fine-tuning base by several Chinese enterprises that want frontier-scale MoE without OpenAI dependency.