GPT-5 cuts political bias by 30%: What does research say

Technology Oct 10, 2025

OpenAI just released GPT-5, calling it their most politically neutral chatbot so far.
Some users miss the "personality warmth" of GPT-4o, but OpenAI's Model Behavior team, led by Joanne Jang, says they've actually cut measurable political bias by 30%—a pretty big step for AI.

Testing the new model on tough political questions

Following their latest developer event, OpenAI unveiled research findings about GPT-5 Instant and GPT-5 Thinking.
Both were tested on 500 prompts—including tough political questions—and showed much more neutral responses than earlier versions.
Researcher Staudacher shared that even under stress tests, political bias stayed only rarely and with "low in severity."
OpenAI admits perfect neutrality is tough, but they're clearly making progress.