GLM-5.1 Achieves First Place in LMArena Code Arena Open Source Rankings, Third Globally

According to monitoring by 1M AI News, the globally authoritative AI evaluation platform LMArena (with blind testing involving millions of users) updated its Code Arena specialized rankings today. GLM-5.1 claimed the top spot among global open-source models and ranked third among all models worldwide. GLM-5.1 not only inherits the open-source SOTA (state-of-the-art) coding capabilities of its predecessor but also achieves breakthroughs in long-horizon tasks, accomplishing: 1. Building a Linux desktop from scratch in 8 hours; 2. Breaking the vector database optimization bottleneck through 655 iterations; 3. Optimizing real machine learning model workloads with 1,000 rounds of tool invocation. Notably, under the same evaluation standards of the METR rankings, GLM-5.1 is the only open-source model capable of sustained 8-hour-level operation. Globally, it is one of the few models with this capability, alongside Claude Opus 4.6.