Political-bias benchmark for Grok 4, GPT-4.1, Gemini 2.5 Pro and Claude Opus 4 | Dark Hacker News