Tencent has released its AI model 'Hy3' as an open model, claiming it is comparable to GLM-5.2 and DeepSeek-V4 at 295B and surpasses GPT-5.5 in scientific tasks.
Tencent, a major Chinese technology company, unveiled its AI model ' Hy3 ' on July 6, 2026. Compared to open models such as GLM-5.2 and DeepSeek-V4-Pro, Hy3 is a smaller model, but it is touted as delivering performance on par with much larger models.
Introducing Hy3
GitHub - Tencent-Hunyuan/Hy3: Hy3 (295B A21B), a leading reasoning and agent model in its size, with great cost efficiency. · GitHub
https://github.com/Tencent-Hunyuan/Hy3
Hy3 is an updated version of Hy3 preview , which was released in April 2026. It is a MoE model with a total of 295 billion parameters and 21 billion active parameters, and is also characterized by having an 'MTP layer' with 3.8 billion parameters.
The benchmark results for 'Hy3,' 'Hy3 preview,' 'GLM-5.2,' 'Seed2.1 Pro,' 'DeepSeek-V4-Pro,' 'Qwen3.7 Max,' 'GPT-5.5,' and 'Claude Opus 4.8' are shown below. Hy3 recorded scores comparable to high-performance open models such as GLM-5.2 and DeepSeek-V4-Pro, and outperformed GPT-5.5 in FrontierScience-Olympiad , which measures the ability to perform scientific research tasks. With a total number of parameters of 295 billion for Hy3, 753 billion for GLM-5.2, and 1.6 trillion for DeepSeek-V4-Pro, it is clear that Hy3 can deliver high performance despite its relatively small size.
Hy3 implements data cleaning and training constraints based on the principle of 'answering only when there is evidence, and clearly stating when there is no evidence, to prevent confusion of information sources and data fabrication,' which significantly reduces 'conflict of facts,' 'fabrication,' and 'logical inconsistencies.' In addition to benchmark tests, it has also been highly evaluated in actual use cases, with the development team of Tencent's AI agent ' WorkBuddy ' reporting that 'when document processing was performed with Hy3 and GLM-5.2, Hy3 completed the processing with 47.4% fewer tokens than GLM-5.2.'
Hy3 is released as an open model at the following link. It is licensed under the Apache License 2.0 .
tencent/Hy3 · Hugging Face
https://huggingface.co/tencent/Hy3
Tencent also offers an API service for Hy3. The fees per million tokens are 1 yuan (approximately 23.9 yen) for input, 4 yuan (approximately 95.4 yen) for output, and 0.25 yuan (approximately 6.0 yen) for cached input. OpenRouter is offering the API for free for a limited time of two weeks from its release.
Hy3 (free) - API Pricing & Providers | OpenRouter
https://openrouter.ai/tencent/hy3:free
Related Posts:
in AI, Posted by log1o_hf