Tencent Hunyuan 3.0 officially launched, focusing on enhancing task execution capabilities.

```

Nearly half a year after joining Tencent, Tencent's Chief AI Scientist Yao Shunyu has released his first report card—Hy3 official version (Hunyuan 3.0 official version).

On July 6, Wallstreetcn noticed that Tencent's AI assistant Yuanbao has launched the official version of Hy3. It has been 74 days since Tencent released the Hy3 preview (Hunyuan 3.0 preview version).

Hy3 is a MoE language model that integrates fast and slow thinking, with a total of 295B parameters, 21B active parameters, and supports up to 256K context length. In terms of parameters, Hy3 is unchanged from the previously released Hy3 preview.

In other words, Tencent insists that Hy3 does not pursue large-scale parameters, but is positioned as "balancing performance and cost-effectiveness," aiming to be one of the best choices for practical scenarios in various business applications.

From Tencent’s perspective, 300B is the optimal balance point between ability and efficiency. Skills like complex reasoning, long-context understanding, and instruction-following are already fully unleashed at this scale. The marginal gains from further increasing parameter size decrease noticeably—doubling the investment only results in single-digit percentage improvements.

Reportedly, building on the Hy3 preview, the official version of Hy3 focuses on optimizing Coding Agent task execution and productivity scenarios, as well as complex reasoning and mathematical capabilities, and fixed some known issues in the preview version.

Meanwhile, Hy3’s pricing has been further reduced. It is currently priced at 1 RMB/million tokens for input, 4 RMB/million tokens for output, and 0.25 RMB/million tokens for input hitting the cache.

Wallstreetcn has learned from Tencent that Hy3 has already been integrated into multiple services under Tencent, including WorkBuddy/CodeBuddy, Yuanbao, Marvis, ima, etc. The API has been launched on Tencent Cloud TokenHub, and several overseas API platforms will soon be connected as well.

Hy3 represents a rhythm adjustment for Tencent in the second half of the AI race.

Since the second half of 2025, Tencent has intensively upgraded the Hunyuan large model team and restructured workflows. In February 2026, the company also rebuilt its large model R&D infrastructure, including pre-training and reinforcement learning, and further improved data quality.

Yao Shunyu once stated that Hy3 preview was the first step in rebuilding the Hunyuan large model.

It is reported that during the rebuilding process of the Hunyuan large model, Tencent established three principles for practical models: First, emphasize systemized capabilities rather than "specialization in one area"; Second, assessment authenticity—actively step outside public leaderboards prone to ranking manipulation; Third, pursue cost-effectiveness.

Therefore, Tencent chose to first launch the Hy3 preview to obtain genuine feedback from the open-source community and users, helping to enhance the practicality of the Hy3 official version.

Data provided by Tencent shows that since its launch, the number of users who independently chose Hy3 preview on WorkBuddy has grown sixfold. In internal evaluation of Hy3 within WorkBuddy office scenarios, task success rate increased from 72% to 90%, and average completion time shortened by 34%.

In the use of Yuanbao, Hy3 addresses the hallucination phenomenon in long text and AI search scenarios through fine-grained data cleaning and training constraints, enabling the model to reliably output under complex evidence.

Reportedly, in tests based on Yuanbao's real log data, Hy3's common-sense error rate decreased by half compared to the preview version, and the hallucination rate dropped by more than half.

Notably, based on Hy3’s greatly enhanced Agent capability, the Agent functionality has also been launched in Yuanbao.

Wallstreetcn found during testing that by clearly asking Yuanbao's task prompt to generate a corresponding file—such as PPT, Word, Excel, PDF, HTML, etc.—the Agent functionality can be triggered, and a complete file can be output, instead of the previous conversational-only output.

Risk Warning and DisclaimerThe market has risks; investment needs to be cautious. This article does not constitute personal investment advice and has not taken into account the unique investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable to their specific situation. Investing based on this information is at your own risk. ```