ChatGPT Upgrades to GPT-5: OpenAI Unveils Its Latest AI Breakthrough

On Thursday, August 7, 2025, OpenAI officially unveiled GPT-5 during an extensive livestream presentation. This new language model has now, and progressively, taken over as the default model in ChatGPT and is also incorporated into the Californian company’s API offerings. Unlike some highly visible technological shifts, GPT-5 represents a structural evolution. According to OpenAI, the system has improved in depth, stability, and adaptability, focusing on transparency and user experience customization.

GPT-5 Deployed Globally, Available in Multiple Formats

GPT-5 is being rolled out worldwide, starting now, though the process may take some time. If available, it is activated by default in ChatGPT for all subscription levels. Free users have access but with limited quotas, while Plus and Pro subscribers enjoy extended capabilities. These subscribers will also have access to GPT-5 Pro, a more powerful version for complex tasks. This widespread deployment allows immediate adoption of the new model, with no prior setup needed.

For developers, OpenAI offers three versions of GPT-5 through its API:

• GPT-5 Standard, intended for complex or intensive uses,
• GPT-5 Mini, which reduces operating costs while maintaining versatility,
• GPT-5 Nano, designed for very low-latency contexts, including embedded or local execution.

This technical segmentation meets concrete needs, including adjusting performance based on usage, controlling infrastructure costs, or enabling integration on resource-limited devices. Additionally, the coding capabilities of GPT-5 are also enhanced compared to its predecessors.

GPT-5 also relies on an intelligent routing system. This internal mechanism automatically determines the best model version for a given query. For instance, a simple factual request will be processed quickly with a lighter model, while a more complex task will activate a deeper thinking mode, without any user intervention. This should make it easier to maintain a balance between speed, cost, and depth of analysis.

“It’s like talking to a real expert, a doctoral-level expert, in any field, on demand,” exclaimed Sam Altman, CEO of OpenAI, during the launch.

A More Cautious AI, Particularly in Sensitive Areas

A major focus of GPT-5, as highlighted during its presentation, is improving the reliability of responses. The model introduces a new feature called safe completions. When faced with a question for which it does not have a reliable answer, GPT-5 will provide a partial response and importantly, explain why it cannot proceed further. If it predicts that it might produce a potentially dangerous response, it will also avoid doing so more effectively than before. This should reduce risks of misinformation, inaccurate claims, or poor advice, especially in sensitive areas like health, law, or finance.

Much effort has also been placed on tone and nuance. In high-stakes contexts, GPT-5 is designed to avoid definitive statements, adopt a more measured style, and highlight the limits of its own capabilities. OpenAI specifies that the model was trained with specific supervision to improve managing uncertainty, consistency of responses over time, and reduction of hallucinations.

A More Customizable Experience in ChatGPT, Both in Tone and Appearance

Beyond technical performance, GPT-5 also introduces advancements in user interaction. OpenAI now offers several ways to personalize the ChatGPT experience, including the addition of “conversational personalities”. Users can choose from four response styles:

• Cynic: a more critical or skeptical tone,
• Listener: an empathetic and attentive style,
• Nerd: a technical, detailed, and sometimes passionate approach,
• Robot: a neutral, concise, and factual formulation.

These personalities are designed not to influence the substance of the responses (the facts or data) but only to alter how they are phrased. According to OpenAI, this new feature aims to better align the AI with user preferences for tone, conciseness, or lack thereof.

The visual appearance is also customizable, with various colors to choose for chat bubbles. The interface adapts to the content, making interactions more dynamic, with options to click on certain passages for additional details or clarification.

What GPT-5 Offers Now (and What’s Yet to Come)

GPT-5 continues to include all the multimodal functionalities of GPT-4o. Handling files such as PDFs, improved and smoother image analysis, and image generation should be accessible. The advanced vocal mode is also improved, promising more natural-sounding voices. For Plus, Pro, and Teams subscriptions, GPT-5 will also be able to connect to Google accounts (Calendar, Gmail, etc.) to better personalize the user experience through its long-term memory capabilities.

Despite its advancements, GPT-5 is not yet a system of AGI (general artificial intelligence), meaning a model capable of performing or learning any cognitive task at least as well as a human. “This is clearly a model with general intelligence, but I think, according to how most of us define AGI, we’re still missing something important, perhaps several important things,” Sam Altman recently explained at a press conference. He noted that GPT-5 represents “a significant step” in that direction.