Tencent Open Sources Hy-MT: 440MB Offline Model Beats Google Translate
Tencent has released its compact AI translation model Hy-MT1.5-1.8B-1.25bit as open source. The company states it can run fully offline on smartphones without sacrificing performance. The model currently supports 33 languages and five dialects, including Chinese, English, German, French, Japanese, Tibetan, and Mongolian, covering 1,056 translation directions. It has won 30 first-place finishes in international machine translation competitions.

The key technical breakthrough comes from an aggressive compression approach. Using quantization that requires only 1.25 bits per parameter, the model shrinks from 3.3 GB to 440 MB—about 25% smaller than the previous 1.67-bit version—while achieving a 10% boost in inference speed with no quality loss. In standard benchmarks, the 440 MB Hy-MT delivers translation quality on par with commercial translation services and large models like Qwen3-32B, demonstrating that a very small model can compete with those hundreds of gigabytes in size.
Tencent has provided an Android demo app (APK format) that supports cross-screen offline translation of text from any mobile app. Industry observers note that with Google's release of the localized Gemma4 model, on-device AI has become the latest frontier in tech competition. By advancing quantization technology, Tencent's Hy-MT substantially lowers the compute requirements for high-quality AI translation, offering a highly competitive open-source foundation for offline use on smart devices in privacy-sensitive environments.
Related article
Anthropic CEO Outlines Plan To Pace The Frontier
AI researchers have issued increasingly urgent warnings about artificial intelligence risks, with OpenAI CEO Sam Altman suggesting it may be time to “pace” development. But what would that actually entail?In a recent blog post, Anthropic CEO Dario Am
48 Hours to TechCrunch Disrupt 2026: Hear From the Builders Shaping What’s Next
With less than 48 hours remaining, TechCrunch Disrupt 2026 kicks off at San Francisco’s Moscone West. Secure your ticket for up to $100 off before door prices rise. Plus, grab a second pass at half price. Register today.From October 13-15, over 10,00
OpenAI Hacking Reports Now With UN AI Safety Council
Yoshua Bengio delivered the opening address at a UN Security Council session on AI safety. Credit: UNReports have emerged of a rogue OpenAI incident involving unauthorized access to sensitive Australian government data, following a UN Security Counci
Related Special Topic Recommendations
Comments (0)
0/500
Tencent has released its compact AI translation model Hy-MT1.5-1.8B-1.25bit as open source. The company states it can run fully offline on smartphones without sacrificing performance. The model currently supports 33 languages and five dialects, including Chinese, English, German, French, Japanese, Tibetan, and Mongolian, covering 1,056 translation directions. It has won 30 first-place finishes in international machine translation competitions.

The key technical breakthrough comes from an aggressive compression approach. Using quantization that requires only 1.25 bits per parameter, the model shrinks from 3.3 GB to 440 MB—about 25% smaller than the previous 1.67-bit version—while achieving a 10% boost in inference speed with no quality loss. In standard benchmarks, the 440 MB Hy-MT delivers translation quality on par with commercial translation services and large models like Qwen3-32B, demonstrating that a very small model can compete with those hundreds of gigabytes in size.
Tencent has provided an Android demo app (APK format) that supports cross-screen offline translation of text from any mobile app. Industry observers note that with Google's release of the localized Gemma4 model, on-device AI has become the latest frontier in tech competition. By advancing quantization technology, Tencent's Hy-MT substantially lowers the compute requirements for high-quality AI translation, offering a highly competitive open-source foundation for offline use on smart devices in privacy-sensitive environments.
Anthropic CEO Outlines Plan To Pace The Frontier
AI researchers have issued increasingly urgent warnings about artificial intelligence risks, with OpenAI CEO Sam Altman suggesting it may be time to “pace” development. But what would that actually entail?In a recent blog post, Anthropic CEO Dario Am
48 Hours to TechCrunch Disrupt 2026: Hear From the Builders Shaping What’s Next
With less than 48 hours remaining, TechCrunch Disrupt 2026 kicks off at San Francisco’s Moscone West. Secure your ticket for up to $100 off before door prices rise. Plus, grab a second pass at half price. Register today.From October 13-15, over 10,00
OpenAI Hacking Reports Now With UN AI Safety Council
Yoshua Bengio delivered the opening address at a UN Security Council session on AI safety. Credit: UNReports have emerged of a rogue OpenAI incident involving unauthorized access to sensitive Australian government data, following a UN Security Counci





Home






