Hy3 preview

The official tencent/Hy3-preview Hugging Face model card describes Hy3 preview as a 295B-parameter Mixture-of-Experts model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team. It is identified as the first model trained on Tencent's rebuilt infrastructure and the strongest shippe Highlights section details strong performance on challenging STEM benchmarks including FrontierScience-Olympiad and IMOAnswerBench, with strong results on the Tsinghua Qiuzhen College Math PhD qualifying exam (Spring '26) and China High School Biology Olympiad (CHSBO 2025). For context learning and instruction followin

RohitAI's analysis details Tencent's official Hy3 release on July 6, 2026, which followed the April Hy3 preview launch with the same MoE architecture (295B total parameters, 21B active, 256K context window). The article frames Hy3 as a "product-first" model designed for reasoning, coding, agents, office workflows, fina Tencent reports that daily token consumption for Hy3 usage grew 20x across global developers and Tencent's own businesses since the preview phase. The release strategy connects the model directly to Tencent's product ecosystem—including Yuanbao, WorkBuddy, CodeBuddy, ima, Marvis, Tencent Docs, QQ, and eventually WeChat

Tencent Cloud published a technical overview of Hy3 preview as "The First Step of Tencent Hunyuan's Rebuild," covering the April 23, 2026 open-source release. The article specifies Hy3 preview as a hybrid fast-and-slow-thinking Mixture-of-Experts model with 295B total parameters, 21B active parameters, and up to 256K c The Tencent Cloud article reiterates the three principles for practicality established during the February 2026 infrastructure rebuild: systematic capabilities avoiding lopsided models, evaluation authenticity stepping outside easily gamed leaderboards, and cost-effectiveness pursued through deep model-inference co-des

According to Baidu Baike, Tencent's Hunyuan team released and open-sourced Hy3-preview on April 23, 2026, as a Mixture-of-Experts model with 295 billion total parameters, 21 billion activated parameters, and a 256K context window. The model was deployed across multiple Tencent products including Tencent Cloud, Yuanbao, On April 29, 2026, OpenRouter's global large model API call volume ranking placed Hy3-preview first overall, first in tool calling scenarios, and second in programming scenarios. Tencent officially released the successor Hy3 model on July 6, 2026, with further improvements in complex reasoning, instruction following, a