/ Tencent just took its AI global overnight.
Liam Ortiz is a tech journalist who covers AI related big tech and breaking news at TheTweaks. Before joining TheTweaks he worked for almost four years in corporate and national tech news in different companies. Few are quick but Liam is quicker, he breaks news before anyone else and that makes her special. His passion is somewhere connected with profession as his hobby is watching documentary movies.
Tencent has started to roll out its own flagship AI model, Hy3 (previously known as Tencent Hunyuan), to international access in the recent few weeks since its official launch on July 6. The wider rollout also affords developers, everyday users and businesses a couple of extra chances to tinker and innovate on the model, as Tencent seeks to ramp up its worldwide AI rivalry (Tencent).
The biggest change to the regular users is that Hy3 will now also be accessible and available worldwide for free in Tencent’s AI agent workspace WorkBuddy until August 31, 2026 (PT). In addition, the model also integrates with Tencent Design Miora, a creative studio tool and Tencent Cloud TokenHub, a multi model management platform. Tencent reports further integration of its AI driven cloud products are in the works.
Developers can access Hy3 via API, and can integrate into coding extensions, third-party platforms, and existing developer environments, fitting in without a rebuild.
Tencent is happy to boast about how many people have adopted its products. After just one week of its release, the model, Hy3, has received more than 68 times as many API calls as the previous model, and was the number one model on OpenRouter’s global LLM usage leaderboard. This is a huge step and it’s helped the model become popular from the start on third-party developer platforms, such as Hermes, Kilo, Cline, OpenClaw, OpenCode and Cherry Studio, and even open-source developer communities like Hugging Face and ModelScope.
Hy3’s architecture is technically a hybrid fast-and-slow-thinking Mixture-of-Experts (MoE) model that has 21 billion active parameters out of 295 billion total parameters and a context length of up to 256K tokens. It matches the top models with two to five times fewer parameters, for both reasoning and coding tasks and for instruction-following and agentic tasks, Tencent claims.
It’s the same level of efficiency it demonstrates in real-world scenarios like software development, office productivity, financial modelling, front-end design, and game production, where Hy3 is said to be very consistent and reliable, and also cuts back on the amount of token used on the go.
It is interesting to note that Tencent is not only referring to Hy3 as a product that is released and forgotten, but it’s also a model that it has developed as a result of its own products. The feedback from millions of users in heavy duty, large scale settings is used to continually evolve the product. The more the product is used, the better it becomes, the more users are able to enjoy it.
It is echoed in Tencent’s most popular AI agent workspace, WorkBuddy, which enables users to use natural language to perform various research, data analysis, document preparation and work communication tasks with multitasking support for more than 100 pre-installed skills.
If this is the case in independent tests in the future, Tencent has said that Hy3 has increased the success rate of the tasks by more than 90% and shortened the average time to complete the tasks by 34% compared to the previous Hy model, there is good reason to keep an eye on the future.
On the creative side, Miora integrates Hy3 into an AI-powered design studio that will be used by design teams and marketers for a simple input and get a full graphic, video, 3D asset or UI element, but without it being treated like a black box.
Developers can use, modify and deploy Hy3 in their own environment without the hassle of restrictions, as it is licensed under the commercially permissive Apache 2.0 license. When it comes to pricing, Hy3 API access is currently low cost, with $0.1288 per million input tokens and $0.5336 per million output tokens, making it an affordable option for scaling agentic workflows.
Tencent is also looking to strengthen itself in the region, collaborating with South Korea’s Cafe24 and Japan’s Metelix to include its accelerator in their respective AI gateway and platform services, which will be supported by infrastructure in 66 Availability Zones across 23 regions and over 3,200 acceleration nodes globally.
Tencent Cloud International’s SVP Poshu Yeung said that it was an “evolution from experimentation to reliable and scalable deployment with AI”. It will be interesting to see whether or not Hy3 can match that success in the international market, but with its free trial period, competitive pricing, open license terms and solid product integration, it’s one of Tencent’s most ambitious global AI advances to date.
From TheTweaks Point of View, Tencent is not about to enter the competitive AI field with yet another product; it is attempting to make Hy3 irresistible to everyone by offering it for free, with API pricing that’s exceptionally cheap, and open licensing. The claims are controversial (models up to several times the size are available and these claims are free for a trial run until August 31, which is low risk, and easy to try. According to internal data, Tencent’s in-house solution, Hy3, could prove to be a viable option for those developers and teams seeking to expand their AI solutions without the corresponding expense. It’s worth a try until the free window runs out.
Tim Cook’s 15 years as Apple’s chief executive officially ended on September 1 and now the torch has been passed on to John Ternus, who is the company’s long-time hardware. It’s Apple’s first leadership change…
Be respectful and constructive. Have a question or feedback? We’d love to hear from you. Contact us at contact@thetweaks.com