MCP server delegates tasks to GPT-5.6, DeepSeek, GLM, and Qwen, benchmarks them

A Reddit user ran 198 benchmark runs with hidden tests to evaluate model performance via an MCP server. The server lets Claude Code delegate tasks to other models, with results compared against Claude.