Hunyuan-MT-7B and Chimera
7B open translation model covering 33 languages that won first place in 30 of 31 language pairs entered at WMT25, plus an open ensemble model.
The training pipeline runs pretraining, cross-lingual pretraining, SFT, translation RL and an ensemble refinement step (Chimera), which Tencent says is the first open ensemble translation model.
- Date
- Monday, 1 September 2025
- Lab
- Tencent
- Kind
- open-weights
- Access
- open weights (restricted license)
Figures
| Measure | Value | Measured by |
|---|---|---|
| WMT25 language pairs placed first | 30 of 31 | company |
| Languages supported | 33 incl. 5 Chinese minority languages | company |
Open-sourcing date from the model card news line (2025.9.1). 'First' claim is Tencent's.
Sources
This record was checked against its sources on 6 October 2026. How we check