Tencent’s open-source translation model beats Google, OpenAI in top global AI competition
Hunyuan-MT-7B came first in 30 out of the 31 tests in a general machine-translation competition held as part of the coming WMT25 conference
Chinese tech giant Tencent Holdings on Monday released a new open-source translation model that aced a global machine-translation competition despite its small parameter size, highlighting the progress the country has been making in artificial intelligence.
Tencent’s Hunyuan-MT-7B came first in 30 out of the 31 tests in the general machine-translation competition, held as part of the Tenth Conference on Machine Translation, also known as WMT25, according to preliminary results published by the event’s organiser. WMT25 serves as a platform for researchers and developers to showcase their progress in computer-aided translation.
Despite having only 7 billion parameters – a measurement of AI models where bigger is generally better – Tencent’s model punched above its weight, outgunning significantly larger and closed-source models from global competitors including Google and OpenAI.
Shenzhen-based Tencent credited the new Hunyuan model’s translation prowess to a “complete translation model training paradigm”, which covers dataset curation, continual pre-training, and multiple fine-tuning and reinforcement learning – techniques used to tweak models to produce optimal performance.
“Hunyuan-MT-7B’s strength is that it uses a small number of parameters to deliver results that measure up to or even surpass larger models,” the Tencent team, responsible for developing the Hunyuan series of AI systems, said in a statement.