Hy-MT2-30B-A3B

Hy-MT2-30B-A3B

A 30B mixture-of-experts translation model from Tencent that translates among 33 languages and follows translation instructions. For teams that need high-volume machine translation on their own GPUs.

Made by tencent on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Hy-MT2-30B-A3B is the largest model in Tencent's Hy-MT2 family of multilingual translation models. The family has three sizes, 1.8B, 7B and 30B-A3B, and the last one is a mixture-of-experts model. Each size translates among 33 languages and follows translation instructions written in several languages.

Who it suits: Teams that need strong machine translation and can host a 30B model, and researchers who study translation with instructions.

Industry | Who uses it | How they use it | Example
Localization | Translation teams | Translate product text among many languages | Convert an app's strings between Chinese and English
Media | Subtitle and content teams | Translate scripts and captions | Render a video script in French or Japanese
Research | Machine translation researchers | Study instruction-following translation | Test how a model follows style and format instructions

You can skip it if you only have a small GPU. The same family has 1.8B and 7B versions that need far less memory.

Where to use it:

• Python with the transformers library, on a machine with a large GPU
• Windows, macOS or Linux, to download the files
• The family's quantized and GGUF copies, for smaller setups

Use case | Who | What they get back
Multilingual translation | Localization teams | Text translated among 33 languages
Instruction-guided translation | Editors | Translations that follow instructions you write
Benchmarking | Researchers | A model to test on the authors' instruction-following benchmark, IFMTBench

How to set it up: You need Python and a GPU that can hold a model of this size. The weights are about 60.1 GB in BF16 and about 15 GB at 4-bit by estimate. The part of the card used here cuts off before its usage steps, so follow them on the model page.

• Python and pip
• A large GPU, or a quantized copy for less memory
• Disk space for the weights

1. Download the model files.
2. Open the usage section on the model page.
3. Load the model with transformers.

pip install -U huggingface_hub
hf download tencent/Hy-MT2-30B-A3B

How to use it: You give the model text and an instruction about the language and style you want, and it returns the translation. The authors also released a Hy-MT2-Translator skill that makes it easier to plug the models into translation tasks.

Example: You ask it to translate a short English product description into Japanese in a polite tone. What comes back is the Japanese text. The card says it follows instructions of this kind across several languages.

For newcomers and experts: If you are new, try the smaller 1.8B or 7B sibling first, since the 30B model needs a large GPU. Translate one short paragraph and compare it with a result you trust.

For experts, this is a "fast-thinking" model, and the 30B-A3B size uses a mixture of experts with 30.1B parameters. The card claims the 7B and 30B-A3B sizes beat several open-source models in that mode, but the numbers sit in an image that is not readable here. The 1.8B size has a 1.25-bit quantized version that the card says is 440 MB.

Source text and instruction
Hy-MT2 mixture-of-experts model
Translation in the target language

Good to know:

• Family: 1.8B, 7B and 30B-A3B, with FP8 and GGUF copies for some sizes
• Languages: 33 languages are supported, and the record lists Chinese, English, French, Portuguese, Spanish, Japanese, Turkish and Russian
• Claims: quality comparisons come from the authors' own report
• License: Apache 2.0
• Cost to run: no API key or paid service named, but it needs a large GPU
• Data it reaches: downloads files from Hugging Face, nothing else is named
• Not for: small GPUs without a quantized copy

On Hugging Face, 13 November 2026

• Task: translation
• Library: transformers
• Parameters: 30.1B
• Weights: about 60.1 GB in BF16; about 15 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 4,684
• Likes: 518
• License: apache-2.0
• Languages: zh, en, fr, pt, es, ja, tr, ru
• Quantized versions: 26
• Fine-tunes: 1
• Tags: transformers, safetensors, hy_v3, text-generation, translation

Links

• Model: https://huggingface.co/tencent/Hy-MT2-30B-A3B
• Files: https://huggingface.co/tencent/Hy-MT2-30B-A3B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download tencent/Hy-MT2-30B-A3B

Then follow the usage in its model card. Nothing in this kit runs by itself.