Tencent's Hy3 model scored 2.67 vs Zhipu's 2.51 in a blind bake-off with 270 experts. The open-source launch puts Tencent in direct competition with China's frontier labs.
Alpha Score of 47 reflects weak overall profile with poor momentum, moderate value, strong quality, moderate sentiment.
Tencent on July 6 launched Hy3, a 295-billion-parameter model with 21 billion active parameters. The same day it ran a blind head-to-head against Zhipu's GLM-5.1. Two hundred seventy domain experts graded both models on the experts' own real work. Across 312 valid comparisons, Hy3 scored 2.67 on a four-point scale. GLM-5.1 scored 2.51.
The test was run on Zhipu's home turf. GLM-5.1 is widely trusted in China's open AI ecosystem. Tencent had no obligation to run a public blind bake-off. That it did, and won, signals a strategy shift. For years Tencent built models quietly, for internal products, judged by whether they shipped. This launch is a bid to be scored as a frontier lab.
Hy3 is released under Apache 2.0 with a 256K context window. It is live on OpenRouter and Tencent Cloud from day one. The inference bill is on the providers until July 21, 2026, when the free trial ends. The model's performance is sharp enough to change which tasks users route to which provider, not sharp enough to dominate outright. Both facts matter.
The move puts Tencent in direct competition with other Chinese AI labs that have been more aggressive about public benchmarks. The free trial period gives developers a chance to test Hy3 against existing options before committing to a provider. Tencent's TCEHY stock page shows an Alpha Score of 51, reflecting a mixed outlook as the company pivots toward open competition.
Drafted by a large language model from the source reporting linked above, then screened by automated publishing checks. It is not read by a journalist before publication. Some articles cite our Alpha Score. Verify prices and figures against the original source. Educational coverage, not personalized advice.