Tencent releases Hy3, a 295B open-weight MoE model
The 295B-parameter, 21B-active model was released under Apache 2.0 and, Tencent said, matched much larger rivals GLM-5.2 (753B) and DeepSeek-V4-Pro (1.6T) while using far fewer tokens.
- Open weights & ecosystem
- Models & capabilities
- Benchmarks & progress
- Notable
Tencent released Hy3 under an Apache 2.0 licence, a mixture-of-experts model with 295 billion total parameters and 21 billion active per token, a 256K-token context window, plus a smaller 3.8-billion-parameter component for speculative decoding. It followed an April 2026 preview version released under a more restrictive Tencent community licence that had excluded the EU, UK and South Korea; the full release dropped those field-of-use restrictions entirely.
Tencent said Hy3 matched the performance of substantially larger open models — GLM-5.2, at 753 billion parameters, and DeepSeek-V4-Pro, at 1.6 trillion — despite its smaller active-parameter count, and that it outperformed OpenAI’s GPT-5.5 on FrontierScience-Olympiad, a benchmark measuring scientific-research tasks. In document-processing tests it reportedly used 47.4% fewer tokens than GLM-5.2 to complete comparable work. The company said the final release incorporated additional data cleaning and training constraints aimed at reducing fabrication and logical inconsistency compared with the preview.
The release added to a run of large open-weight Chinese MoE models through 2026 — alongside GLM-5.2 and DeepSeek’s line — competing on token efficiency and permissive licensing rather than on raw parameter count, at a point where several Chinese labs’ open releases were being benchmarked directly against the leading closed US models.