Chinese AI Model GLM 5.2 Outperforms Anthropic Claude Code on Cybersecurity Benchmark
GLM-5.2, an open-weight language model released on June 16 by Zhipu AI, scored roughly 39 percent on an independent cybersecurity benchmark measuring IDOR vulnerability detection, outperforming Anthropic's Claude Code, which scored 32 percent. The model, developed at the Tsinghua University spinout and distributed under the MIT license, also ran at approximately one-sixth the cost per vulnerability identified compared to rival US systems.
The performance has drawn comparisons to previous shifts in artificial intelligence development. According to Semgrep's testing and developer analysis, GLM-5.2 ranks among the top three models globally for the specific security task, closing a performance gap where American labs previously held a clear lead. At Hugging Face, where the model's weights and post-training code are freely available, users reported strong institutional demand and noted that the architecture can run on standard open inference stacks and consumer Apple hardware.
From the sources (8 posts)
@algo_diverRT @VukRosic99: GLM 5.2 post-training code is OPEN SOURCE (slime) Megatron-LM trains. SGLang generates the rollouts. A single data buffer…
@clashreportBIG: A Chinese AI model from has caught up with its American counterparts in the field of cybersecurity, according to independent benchmarks. GLM-5.2, released on June 16 under an MIT license by
@polymarketRT @Polymarket: JUST IN: A new Chinese AI model from Zhipu AI reportedly matches Claude Mythos’ performance at finding security bugs.
@kimmonismusThe release of GLM-5.2 is the second DeepSeek moment.
@quixiaiRT @antirez: GLM 5.2, Q2_K routed experts (effectively ~2.6 bits) running with SSD streaming on an M5 Max 128GB computer.
@yuchenj_uwGLM-5.2 is the open-source Claude moment. The demand we’re seeing at Databricks is astonishing. The world is going to see massive adoption of oss LLMs. Also, more companies will shift toward post-training their own models on top of oss mo
@aaditya_aiRT @Yuchenj_UW: GLM-5.2 is the open-source Claude moment. The demand we’re seeing at Databricks is astonishing. The world is going to see…
@catcheronthesly'While other Chinese releases have made significant gains, GLM-5.2 is the first model to rank in the top three globally on a major benchmark.' China’s Zhipu AI sparks new ‘DeepSeek moment’ with cost-effective coding model