Pattern: Open-Weight Models Achieve Frontier Coding Parity — GLM-5.1 Tops Both Opus 4.6 and GPT-5.4 on SWE-Bench Pro

VentureBeat·high signal

GLM-5.1's 58.4 on SWE-Bench Pro (vs Opus 4.6 at 57.3, GPT-5.4 at 57.7) marks the first time an open-weight model leads the most rigorous coding benchmark. Combined with its 8-hour autonomous execution capability and MIT license, this creates a viable self-hosted alternative for agent pipelines currently dependent on API providers. The cost arbitrage is stark: $1.40/M input tokens via API, or free via vLLM/SGLang on your own hardware. Builders running daily agent pipelines should evaluate GLM-5.1 as a primary or fallback model to reduce vendor lock-in.