GLM-5.1 Open Weights SOTA Agentic Coding Benchmarks
Key Questions
What are GLM-5.1 benchmarks?
GLM-5.1 achieves SOTA agentic coding: SWE-Bench Pro 58.4%, VectorDBBench 21.5k QPS, KernelBench 3.6x, Linux desktop 8hrs. Ideal for long-horizon agents.
How to self-host GLM-5.1?
Use vLLM/SGLang for efficient inference, with new courses by Andrew Ng. Supports text/image generation and agentic engineering.
Where is GLM-5.1 available?
GLM-5.1 is live on Poe for agentic engineering, repo gen, terminal workflows, and reliability. Poe enables easy access to its capabilities.
Dev guide long-horizon agents: SWE-Bench Pro 58.4%, VectorDBBench 21.5k QPS, KernelBench 3.6x, Linux desktop 8hrs; vLLM/SGLang self-host + Poe for agentic engineering/repo gen/terminal workflows/reliability.
Sources (2)
Updated Apr 10, 2026