GLM-5.1 is a model with enhanced coding capabilities that achieves state-of-the-art 58.4% on SWE-Bench Pro, outperforming GLM-5, GPT-5.4, Opus 4.6, and Gemini 3.1 Pro.

It sustains performance over long horizons, optimizing a vector database to 21.5k QPS across 600 iterations, delivering 3.6× speedup on KernelBench Level 3 after 1,000+ turns, and building a full Linux-style desktop web app over 8 hours. Unlike predecessors, it iterates by analyzing results, revising strategies, and handling ambiguous tasks without plateauing early.