Open Models have crossed a threshold

Recent evaluations show that open models like GLM-5 and MiniMax M2.7 now perform on par with closed frontier models for core agent tasks such as tool use and instruction following. These open models offer significant benefits in cost and latency, running up to 8-10 times cheaper and with much faster response times compared to proprietary counterparts. The analysis was conducted using the LangChain Deep Agents evaluation harness, which measures correctness, solve rate, and efficiency across various tasks including file operations and retrieval. This performance parity, combined with lower operational costs, makes open models a viable and attractive option for developers building and deploying agentic workflows in production environments.