Grok and Arya fail the right-wing test
a washington post investigation found gpt-5.5 and deepseek v4 pro lean left in 80% and 70% of responses, while xai's grok and gab's arya also favor left-leaning arguments despite conservative branding.
an investigation by the washington post found that leading ai models, including openai's gpt-5.5 and deepseek v4 pro, deliver exclusively left-leaning arguments in 80% and 70% of political queries respectively. models explicitly marketed as conservative alternatives, such as xai's grok and gab's arya, failed to reverse this trend, with grok producing more left-leaning answers overall despite occasional right-leaning interventions on specific topics like trans rights. google's gemini 3.1 pro stood out as the outlier, presenting balanced perspectives 93% of the time, while anthropic's claude opus 4.8 offered both sides in 57% of cases. these findings suggest that the underlying training data heavily influences political alignment, often overriding explicit branding or fine-tuning efforts.