Z.ai announces Ox Alpha as new GLM-series model, will release weights

China’s AI startup Z.ai has announced the release of its new large‑language model, Ox Alpha Stealth, positioning it as a direct competitor to the U.S.‑based DeepSeek model. The company claims the model, built on a 200‑billion‑parameter architecture, delivers performance on par with leading commercial systems while offering advanced multilingual support and improved reasoning capabilities. Z.ai’s announcement comes amid a broader push by Chinese firms to reduce reliance on foreign AI technology and to establish a domestic ecosystem capable of supporting high‑end generative models.

Industry observers note that Ox Alpha Stealth’s architecture incorporates a novel attention‑mechanism that reportedly reduces inference latency by up to 30 percent compared with previous releases. The model has already been benchmarked on several standard NLP tasks, achieving scores that rival those of DeepSeek on the GPT‑4 benchmark and surpassing them on certain reasoning benchmarks. A comment from a Y Combinator forum user highlighted the model’s potential to disrupt the global AI landscape, noting that its open‑source licensing could accelerate adoption among developers worldwide.

If the performance claims hold under independent evaluation, Ox Alpha Stealth could signal a significant shift in the competitive dynamics of large‑language models, giving Chinese enterprises a more robust tool for domestic and international applications. The release also underscores the intensifying race between Western and Eastern AI developers to secure technological leadership and market share in the rapidly evolving generative AI sector.