New GPT-5 model unveiled by ChatGPT.
OpenAI, the leading artificial intelligence research company, has announced the launch of its latest language model, GPT-5. This new model promises substantial improvements and new features over previous versions, particularly in reducing hallucinations and lies, enhancing text generation, medical advice, and formal test performance.
The new GPT-5 model is designed to provide safer responses when discussing complex topics, aiming to make it possible for users to discuss anything with the model, but the responses won't harm either the user or society. One of the key improvements is a reduction in hallucinations by up to 78-84% compared to earlier models like o3, achieved through enhanced factual consistency training, better source attribution, improved handling of uncertainty, and advanced verification.
GPT-5 exhibits smarter reasoning, including control over the amount of reasoning effort applied (minimal to high) and answer verbosity (low to high), enabling more accurate and context-appropriate text generation for complex or simpler tasks. This flexibility helps in tailoring responses to user needs efficiently.
In the medical field, GPT-5 is notably more accurate and self-aware in handling health-related questions, achieving 46.2% on the challenging HealthBench Hard benchmark. It is trained to be more cautious and precise with medical information, although users are still advised to verify outputs as with all AI models.
GPT-5 sets new state-of-the-art results in several academic and benchmark evaluations, including math, coding, multimodal understanding, health, and general knowledge. These gains exceed those of comparable prior versions, reflecting its stronger reasoning, instruction following, and multimodal capabilities.
The new model also introduces a unified system combining text, images, voice, and live video inputs, supports dynamic routing for better responses, integrates tightly with developer tools, and offers real-time coding and guidance. It supports vast context windows, enabling handling of lengthy documents or conversations.
Users can choose from three options: ChatGPT 5 (for everyone), GPT-5 Thinking (for complex tasks, available for paid users), and GPT-5 Thinking Pro (for pro users only). In quick mode, the AI provides a shorter and faster response without deep thinking. In complex ambiguous test scenarios, the number of lies for GPT-5 has halved compared to the 30 model.
The voice mode has been updated and supports custom GPTs. Paid users can adjust the voice to their liking. Users will soon be able to choose a color for some interface elements like the Voice button, highlighted text, and chat bubbles. In the "Personalization" settings, users can choose from four characters (Cynic, Robot, Listener, and Nerd) or stay in the standard ChatGPT style.
OpenAI's CEO stated that GPT-5 "has reached the level of a PhD." The new model can decide when to answer quickly and when to think longer, a feature developers call the "router". GPT-5's medical advice capabilities are highlighted, with OpenAI offering ChatGPT not as a replacement for a doctor, but as a partner for understanding results and decision-making.
In summary, GPT-5 offers the most powerful and reliable step forward in OpenAI’s language models to date, with major improvements in truthfulness, reasoning, text and medical advice quality, and test performance, along with expanded multimodal and developer-friendly features. Users still should verify outputs in high-stakes settings, but the reductions in hallucinations and lying represent a dramatic quality-of-life advance.
The GPT-5 model, a significant advancement from OpenAI, is equipped with artificial-intelligence capabilities, designed to provide safer and more accurate responses when discussing complex topics, especially in reducing lies and enhancing text generation, medical advice, and formal test performance.
GPT-5's artificial-intelligence enables it to exhibit smarter reasoning, offering control over the amount of reasoning effort applied and answer verbosity, thereby providing context-appropriate text generation for complex or simpler tasks.
