Zhipu AI has released GLM-5.3, claiming it as the strongest open-weights coding model with 50% performance improvement over its predecessor. The model weights will become open source in two weeks.
GLM-5.3 achieves its performance gains through post-training optimization alone, according to Zhipu AI's benchmarks. The model has been specifically trained for cybersecurity applications, demonstrating practical utility by identifying 2,436 vulnerabilities across 269 projects during testing.
The release positions GLM-5.3 as a competitive alternative to proprietary coding models, expanding the landscape of open-source AI tools available to developers and security teams. By making weights publicly available, Zhipu AI enables researchers and organizations to deploy the model locally without reliance on external APIs.
The cybersecurity focus reflects growing demand for AI-assisted vulnerability detection as security teams face resource constraints. GLM-5.3's track record of finding thousands of vulnerabilities suggests practical applications in code review and security auditing workflows.
Open-weights models have gained traction as organizations seek greater control over their AI infrastructure and reduce dependency on closed platforms.
Bumble has abandoned its defining feature—requiring women to message first—as online dating platforms struggle with user fatigue and dwindling engagement. The move signals a broader industry shift toward AI solutions as traditional swipe-based models falter.
Republican state lawmakers are drafting legislation to legally prohibit marriages between humans and chatbots. The measures aim to preempt a practice that has already gained traction among some users.
David Reger, CEO and founder of German robotics company Neura, says the velocity of physical AI development is extraordinary and positions Europe to capitalize on a new growth sector.