Wha is latest from Chinese LLM model launch and performance?

The latest Chinese LLM launch is GLM-5 from Z.ai, released in mid-February 2026 as a 744-754 billion-parameter open-source model under an MIT license [4][5][12]. Trained entirely on domestic Chinese chips like Huawei Ascend, it achieves 80-90% of frontier-model performance and scores competitively on benchmarks, outperforming models like Gemini 3 Pro and Grok 4 while rivaling Claude Opus 4.6 and GPT-5.2 [5][12]. It's available via API at a low cost of $1 per million input tokens, aiming to undercut Western providers and enhance China's AI self-sufficiency [4][5][12]. Benchmarks show impressive results overall, but the system card highlights gaps compared to US closed-source models in code generation and broad knowledge areas [1]. This positions GLM-5 as a cost-effective alternative that could pressure US firms' margins in the global AI market [4][5][12].