According to 1M AI News monitoring, today the renowned global AI benchmarking platform LMArena (with one million users participating in a blind test) updated the Code Arena specialized ranking. GLM-5.1 claimed the top spot in the global open-source models list and ranked third in the global models list.
GLM-5.1 not only inherits the open-source SOTA coding capabilities of the previous generation of models but also made breakthroughs in Long-Horizon Tasks, achieving:
1. Building a Linux desktop from scratch in 8 hours;
2. Breaking the vector database optimization bottleneck in 655 iterations;
3. Optimizing real machine learning model workloads with 1000 rounds of tool invocation.
It is worth mentioning that under the same evaluation criteria as the METR ranking, GLM-5.1 is the only open-source model that achieves sustained work at the 8-hour level and is one of the few models worldwide with this capability, apart from Claude Opus 4.6.
