Tencent has released Hy3, a 295-billion parameter language model available under the Apache 2.0 open-source license. The company claims the model matches the performance of Zhipu's GLM-5.1 and GLM-5.2.
The release follows Hy3's preview launch in April. By open-sourcing the model, Tencent joins a growing trend of Chinese AI labs releasing large-scale models with permissive licensing, contrasting with many Western counterparts that maintain stricter commercial restrictions.
The 295B parameter scale positions Hy3 among the larger open-weight models available. Tencent's performance claims against GLM-5.1 and 5.2 signal competitive capability in benchmarks and real-world applications.
The Apache 2.0 license permits commercial use, modification, and distribution with minimal restrictions, making Hy3 accessible for researchers and companies building derivative applications.
This move addresses what analysts have noted as a competitive advantage for Chinese AI developers: the willingness to release powerful models openly while Western labs often restrict access through API-only deployments or licensing terms.
Mark Zuckerberg announced that Meta's Watermelon model and open weights version of Muse Spark are coming soon, following the rollout of Muse Spark 1.3 with significant improvements in coding and agentic capabilities.
New York City public schools have prohibited the use of generative AI tools for students through eighth grade, implementing a year-long restriction on the technology in classrooms.
Meta has unveiled its latest artificial intelligence model, with the company's chief AI officer stating its capabilities now rival those of leading competitors in the space.
Anthropic has released a verification tool allowing users to check whether files were created by Claude. The service addresses growing concerns about AI-generated content authentication.