I tested GLM-5.1 — it beat GPT-5.4 & Claude Opus 4.6 and is 7.8× cheaper.

Editor's take

A new open-source model, GLM-5.1, has demonstrated superior performance on SWE-Bench Pro, a benchmark for real-world software engineering tasks, exceeding GPT-5.4 and Claude Opus 4.6, while also offering a significant cost advantage.

This development is noteworthy because it challenges the dominance of proprietary models in complex domains like software development. The accessibility and affordability of GLM-5.1 could democratize advanced AI capabilities for smaller organizations and individual developers, potentially accelerating innovation in the open-source AI ecosystem.

Future attention should focus on the sustainability of this performance advantage. Independent verification of GLM-5.1's benchmark results and its ability to generalize to other complex coding tasks, beyond SWE-Bench Pro, are crucial. Furthermore, understanding the specific architectural or training innovations that enabled this leap will be key to predicting its long-term impact.