A large model that was open-sourced less than two weeks ago has extended its free trial period from two weeks to nearly a month, all because of user feedback saying "another two weeks." On July 20, Zhang Jun, the public relations director of Tencent, revealed on Weibo that when the two-week free trial of the HuanYuan large model Hy3 was about to end, the company received strong feedback from users requesting an extension. As a result, Tencent decided to extend the limited-time free offer for Hy3 to August 5th for WorkBuddy and CodeBuddy users.

image.png

The main focus of this free trial, Hy3, is the new generation model that Tencent officially open-sourced and released on July 6th. Its most distinctive feature is a model that combines fast and slow thinking - it can provide intuitive responses quickly and also take time to perform in-depth reasoning, with both rhythms integrated into the same architecture. In terms of technical foundation, Hy3 uses a MoE (Mixture of Experts) architecture, with a total parameter scale of 295B and activated parameters of 21B, which means that only a small part of the experts are activated during each inference, using relatively modest computing costs to leverage a vast knowledge capacity; it supports up to 256K context length, allowing it to absorb long documents or entire codebases without losing the previous context.