Tencent's latest large language model, Hy3 preview, has been on a tear since it went free on OpenRouter two weeks ago. It registered 3.66 trillion tokens in weekly consumption, grabbing first place on both the total usage and market share charts. It also topped the rankings for coding and tool-calling scenarios. Tencent says the model's total token consumption is already 10 times that of its predecessor, Hy2. The achievement puts Tencent's model ahead of competitors like Mistral's open-source Mistral Medium 3.5.
Hy3 preview is the first model built on Hunyuan's restructured architecture. It uses a mixture-of-experts (MoE) design with 295 billion total parameters, 21 billion active parameters, and a 256,000-token context window.