Leonid Herasimau's Avatar
Leonid Herasimau
AI·last month

OpenAI Unveils GPT-5 — It Will Be Available to Free Users, Too

The company calls it its "fastest, smartest, and most helpful model."

  • According to the developers, it not only surpasses previous models in benchmarks, responds faster, and 'hallucinates' less, but is also more useful for everyday queries.
  • OpenAI says that GPT-5 is the company's most powerful model for developers. It allows for the creation of beautiful and adaptive websites, applications, and games with a single prompt.
In the prompt, the model was asked to create a single-page drawing application with a canvas, tools, a palette, and a grid. Source: OpenAI
In the prompt, the model was asked to create a single-page drawing application with a canvas, tools, a palette, and a grid. Source: OpenAI
Comparison of GPT-5, o3, and GPT-4o on several benchmarks. Source: OpenAI
Comparison of GPT-5, o3, and GPT-4o on several benchmarks. Source: OpenAI
  • Another task the model handles well is generating texts with literary depth and rhythm. But it is also suitable for everyday tasks like drafting reports or emails.
A comparison of how GPT-4o and GPT-5 handled the task of writing a poem. Source: OpenAI
A comparison of how GPT-4o and GPT-5 handled the task of writing a poem. Source: OpenAI
  • The third area of application that OpenAI highlights is medical-related queries. ChatGPT provides more accurate and reliable answers, taking into account the user's context, knowledge level, and location. However, the company noted: the chatbot is not a substitute for a doctor—it is more of a partner that can help interpret results and ask questions.
Comparison of GPT-5, o3, and GPT-4o on the HealthBench benchmark, which evaluates the model's responses to medical queries. Source: OpenAI
Comparison of GPT-5, o3, and GPT-4o on the HealthBench benchmark, which evaluates the model's responses to medical queries. Source: OpenAI
  • GPT-5 uses fewer unnecessary emojis and responds more subtly and thoughtfully compared to GPT-4o. The company believes that communicating with the chatbot should feel less like talking to an AI and more like talking to a friend with a PhD-level intellect. OpenAI noted that it has addressed the issue of sycophancy, which GPT-4o accidentally suffered from.
Comparison of the response accuracy of OpenAI models on 'multimodal benchmarks.' Source: OpenAI
Comparison of the response accuracy of OpenAI models on 'multimodal benchmarks.' Source: OpenAI
Comparison of the response accuracy of OpenAI models in the 'Humanity's Final Exam' test, created to assess AI capabilities in academic questions. Source: OpenAI
Comparison of the response accuracy of OpenAI models in the 'Humanity's Final Exam' test, created to assess AI capabilities in academic questions. Source: OpenAI
  • For the most complex tasks, OpenAI is releasing GPT-5 Pro, which will replace o3-pro. This version of the model thinks longer but provides the highest quality and most comprehensive answers.
  • GPT-5 will be available to free users and Plus, Pro, and Team subscribers starting August 7, 2025. Enterprise and Edu subscribers will get access a week later.
  • Pro subscribers will get unlimited access to GPT-5 and access to GPT-5 Pro. Plus users will have a higher usage limit than free users. Free users will be automatically switched to GPT-5 mini upon reaching the limit.

GPT-5 teaser. Source: OpenAI

#news #openai

0
43
Comments
Please log in to add comments.
Loading comments...
Recommendations