OpenAI has released new research showing that GPT-5 models are 30% less politically biased compared to previous versions, based on testing with 500 prompts covering politically sensitive topics and discussions.
The testing included a wide range of questions — from those with a “liberal tilt” to “conservative” ones — and evaluated responses across five different bias metrics. GPT-5 performed best in emotionally charged situations, although all models, including GPT-5, showed slightly higher bias when responding to strongly liberal prompts.
According to internal analyses, fewer than 0.01% of real ChatGPT conversations exhibit political bias. Researchers identified three main patterns: models sometimes express political opinions as their own, favor only one perspective, or amplify the user’s emotional tone.
This research shows progress but also serves as a warning — even subtle biases in models used by hundreds of millions of people can shape perceptions of reality. Particularly sensitive are moments when users are emotionally engaged, as AI can subtly reinforce existing views or influence opinions.
In brief: Tech World Highlights
• Opera has launched Neon, a new AI browser capable of autonomously performing actions on behalf of users, available through a premium waitlist subscription.
• Meta has acquired Rivos, a chip development startup, aiming to accelerate in-house AI chip production and reduce reliance on Nvidia.
• According to The Information, OpenAI generated 4.3 billion dollars in revenue during the first half of 2025 but spent 2.5 billion on research and computing costs.
• OpenAI’s new social app, Sora, has surged to third place on the Apple App Store — right behind Google Gemini and ChatGPT — following a viral invite-only launch.
• Hume AI has launched Octave 2, a new multilingual text-to-speech model supporting 11 languages and featuring voice conversion and phoneme editing capabilities.
AI Trending Tools:
• Apps SDK – enables users to chat and directly create applications within ChatGPT.
• Hunyuan-Vision-1.5-Thinking – Tencent’s advanced vision-language model.
• PromptSignal – a tool that visualizes how large language models (LLMs) rank your brand.
