Open Models Like GLM-5 and MiniMax M2.7 Match Closed Frontier Models in Agent Tasks

Initial evaluations using Deep Agents show open weight Large Language Models, specifically GLM-5 (z.ai) and MiniMax M2.7, perform similarly to closed frontier models on core agent tasks such as file operations, tool use, and instruction following. Open models offer significant advantages in cost and latency; for instance, MiniMax M2.7 is substantially cheaper than Claude Opus 4.6, potentially saving around $87k annually for a 10M token/day application. Furthermore, GLM-5 on Baseten averages 0.65s latency and 70 tokens/second, significantly faster than Claude Opus 4.6 at 2.56s and 34 tokens/second, making them viable for production workflows.