Model Details
Hy-MT2-7B
Quick answer: Hy-MT2-7B is the mid-size member of Tencent Hunyuan's Hy-MT2 family — a 7B-parameter, Apache 2.0-licensed multilingual translation-specialist model supporting 33 languages. It scores 86.89 (XCOMET-XXL) on FLORES-200's full translation matrix and actually leads the entire Hy-MT2 family — including the larger 30B-A3B — at 63.86 on WMT25's deliberately hard, low-saturation test set.
At a Glance
Where Hy-MT2-7B leads
- Best-in-family score on WMT25 (63.86 XCOMET-XXL), edging out the larger 30B-A3B MoE variant (62.89) — the only benchmark where the mid-size dense model outperforms its bigger sibling
- Strong efficiency-to-quality ratio: dense 7B architecture with straightforward deployment (no MoE routing complexity) while still competitive with much larger general-purpose LLMs on translation-specific evaluation
- Outperforms Tower-Plus-72B (a 10x larger open translation-specialist model) on FLORES-200 and WMT25
Where it lags
- Behind Hy-MT2-30B-A3B on FLORES-200 (86.89 vs 87.47) and on domain/real-world translation benchmarks
- Behind frontier general-purpose models like Gemini 3.1 Pro on raw translation quality across the board
- General instruction-following (IFBench, IFEval, MaXIFE, Multi-IF) is weaker than translation-specific instruction following — it's tuned for translation, not general-purpose chat
Best for: Deployments wanting the best balance of translation quality and dense (non-MoE) architectural simplicity — particularly strong on harder/fresher test material like WMT25.
What Hy-MT2-7B Is
Hy-MT2-7B is the mid-size member (of 1.8B, 7B, 30B-A3B) in Tencent Hunyuan's second-generation "fast-thinking" multilingual translation model family, described in the technical report "Hy-MT2: A Family of Fast, Efficient and Powerful Multilingual Translation Models in the Wild." It's a dense transformer trained via a three-stage Family-Centric Post-training (FCPT) pipeline: MT-oriented mid-training on ~1T tokens of translation-related data, followed by Reference-Guided On-Policy Distillation and Family-specific RL training organized around language families (Western European, East Asian, Middle Eastern right-to-left), then a final Cross-family distillation step into a unified model.
Like its siblings, Hy-MT2-7B is evaluated across four dimensions: general translation (FLORES-200, WMT25, Mandarin⇔Minority), real-world business scenarios (Tencent's own WildMTBench), domain-specific translation (Tencent's own DomainMTBench, covering finance/law/medicine/technology/politics/education), and translation instruction-following (IFMTBench, plus general instruction benchmarks IFBench/IFEval/MaXIFE/Multi-IF). Its standout result is topping the entire Hy-MT2 family on WMT25 — a test set explicitly curated to be hard and resist saturation — suggesting the 7B size may generalize better to genuinely novel source material than the larger MoE variant's higher-capacity-but-more-specialized routing.
Specifications
| Field | Value |
|---|---|
| Organization | Hungyuan (Tencent) |
| Architecture | Dense transformer |
| Parameters | 7B |
| Context length | 262,144 tokens |
| Supported languages | 33 |
| License | Apache 2.0 |
| Release date | May 2026 |
| Modality | Text |
| Quantized variants | FP8, GGUF (multiple quant levels) |
Pricing
Open weights under Apache 2.0 — self-hosted via transformers, vLLM, SGLang, or llama.cpp (GGUF builds available). No API list price; cost is your own inference infrastructure.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| FLORES-200 | 86.89 (XCOMET-XXL) | Hy-MT2 technical report | 2026-05 |
| WMT25 | 63.86 (XCOMET-XXL) | Hy-MT2 technical report | 2026-05 |
Hy-MT2-7B also reports scores on Tencent's own in-house evaluation suites (WildMTBench, DomainMTBench, IFMTBench, Mandarin⇔Minority testset) and on general instruction-following benchmarks (IFBench, IFEval, MaXIFE, Multi-IF) — not added as Benchgen leaderboard entries since these either lack an independent public benchmark page (WildMTBench/DomainMTBench/IFMTBench are Tencent-constructed, non-independently-reproducible test sets) or measure general instruction-following rather than translation capability specifically.
Hy-MT2-7B vs Alternatives
| Model | FLORES-200 (XCOMET-XXL) | WMT25 (XCOMET-XXL) | License | Params |
|---|---|---|---|---|
| Hy-MT2-1.8B | 79.77 | 50.30 | Apache 2.0 | 1.8B |
| Hy-MT2-7B | 86.89 | 63.86 | Apache 2.0 | 7B |
| Hy-MT2-30B-A3B | 87.47 | 62.89 | Apache 2.0 | 30B (3B active) |
Hy-MT2-7B sits between its siblings on overall translation quality, but its WMT25 lead over the larger MoE variant makes it worth considering specifically for harder or less-templated source material.
Related Models & Benchmarks
- Compare: Hy-MT2-1.8B · Hy-MT2-30B-A3B
- Benchmarks: FLORES-200 · WMT25
Frequently Asked Questions
What is Hy-MT2-7B?
Hy-MT2-7B is Tencent Hunyuan's mid-size translation-specialist model, a 7B-parameter dense transformer supporting 33 languages, released under Apache 2.0.Is Hy-MT2-7B open source?
Yes — it's released under the Apache 2.0 license with weights on Hugging Face and ModelScope, including FP8 and GGUF quantized variants.Does Hy-MT2-7B beat the larger Hy-MT2-30B-A3B model?
Only on WMT25 (63.86 vs 62.89 XCOMET-XXL) — on FLORES-200 and the family's other evaluations, the larger 30B-A3B MoE model leads overall.Specs from Tencent Hunyuan's official Hy-MT2-7B Hugging Face model card and technical report (arXiv:2605.22064). Last updated 2026-08-31.