Back to today’s briefing

Story perspectives

GPT-5 Models Slash Political Bias by 30%, OpenAI Claims

10/11/2025

50 9

Summary

Story summary

OpenAI states that GPT-5 Instant and GPT-5 Thinking reduce political bias by 30% versus prior versions.
An internal report shows the models were tested on 100 politically charged topics, with bias assessed by another AI.
Findings indicate bias appears in less than 0.01% of responses, mainly to strongly liberal prompts.
OpenAI says it remains neutral and invites evaluation of its bias framework, while experts warn neutrality may be unrealistic.

1 of 1

Story summary

OpenAI states that GPT-5 Instant and GPT-5 Thinking reduce political bias by 30% versus prior versions.
An internal report shows the models were tested on 100 politically charged topics, with bias assessed by another AI.
Findings indicate bias appears in less than 0.01% of responses, mainly to strongly liberal prompts.
OpenAI says it remains neutral and invites evaluation of its bias framework, while experts warn neutrality may be unrealistic.