Posts tagged with "minimax-m2.5"

I built a sycophancy benchmark for my local AI agent and compared it to frontier models. The biggest surprise wasn't the models — it was how much a simple behavioral layer changed honesty-under-pressure.