OpenAI is reigniting the race with GPT-5
The announcement had been anticipated for over a year. On August 20, 2025, OpenAI officially unveiled GPT-5, the new version of its flagship language model, integrated into ChatGPT and the OpenAI API. This release comes amid fierce competition, as Anthropic’s Claude 3.5, Google’s Gemini 2, and DeepSeek-Vision are rapidly transforming the landscape of generative artificial intelligence.
Following GPT-4 (March 2023)and GPT-4o (May 2024), this new iteration does more than just boost raw processing power. It marks a strategic turning point in how an AI model understands, remembers, and adapts to its users. OpenAI presents GPT-5 as its most reliable, consistent, and aligned model to date1.
What GPT-5 Actually Changes
GPT-5 significantly improves several key areas:
- Advanced contextual understanding: The model is better able to follow complex or lengthy instructions, even in the middle of a conversation.
- More nuanced and better-structured answers: summaries are more relevant, references are cited more accurately, and reasoning is more detailed.
- Enhanced conversational context tracking: GPT-5 retains a record of past exchanges within the same session with greater consistency than its predecessors.
One of the major improvements is the long-term memory feature, which can be enabled in ChatGPT. It allows the assistant to remember information from one session to the next (name, preferences, writing style, etc.), and users can view, edit, or delete this information2.
A more robust and specialized architecture
OpenAI has not disclosed the details of GPT-5’s architecture, but several clues suggest it is a composite model similar to the Mixture of Experts approach:
- Specialization of submodels by task: mathematics, code, summarization, logic…
- Dynamic activation of experts based on the query, optimizing resources and accuracy.
- Optimized for long-form generation: GPT-5 excels at handling lengthy texts, technical reports, and complex arguments.
In terms of multimodality, GPT-5 remains primarily text-based at this stage, but it integrates with ChatGPT through the vision, voice interaction, and document processing capabilities provided by GPT-4o. OpenAI has already hinted at an upcoming native multimodal version, though no official release date has been set.
Deployment and accessibility
GPT-5 is now available through several channels:
- ChatGPT (web or mobile version), for subscribers to the ChatGPT Plus or Enterprise plan
- OpenAI API, accessible via the developer console or via Microsoft Azure
- Primary language: English, with support for French, Spanish, German, Japanese, etc. to be added gradually.
Users can enable or disable personal memory in the settings and view what the model "knows" about them. This feature is designed to provide a more personalized experience, particularly for professional or educational use.
Ethical Issues and Limits
Despite its impressive performance, GPT-5 raises several fundamental questions:
- Cross-session data retention raises privacy concerns, despite the available control options. Who is liable in the event of a data breach or storage error?
- The realistic nature of the responses can mask hallucinations, which are harder for non-expert users to detect.
- Using the model in sensitive contexts (healthcare, law, finance) requires more robust safeguards.
OpenAI claims to have strengthened its evaluation protocols (red teaming, alignment, user feedback), but the challenges associated with aligning AI at scale remain fully intact3.
GPT-5: Toward a General-Purpose AI Assistant?
With GPT-5, OpenAI is taking a step toward creating a universal assistant: one capable of understanding complex contexts, engaging in long-form conversations, adapting to different roles (coach, writer, translator, coder, etc.), and interacting with external tools.
But the company remains cautious in its communications. It avoids terms like AGI (Artificial General Intelligence) and emphasizes a gradual, controlled, and responsible approach.
In an increasingly competitive market, GPT-5 solidifies ChatGPT’s position as the leading AI interface, while laying the groundwork for future developments: proactive assistants, autonomous agents, and AI integrated into everyday life.
Learn more
To learn more about the AI revolution in visual and multimodal search, read this accessible and complementary article: Perplexity Unveils Comet, an AI-Powered Browser
This content explores how AI integrated into a browser is transforming access to information—a dynamic similar to the generative AI now incorporated into Google Search.
References
1. OpenAI. (2025). Introducing GPT-5.
https://openai.com/research/gpt-5
2. OpenAI Help. (2025). About Memory in ChatGPT.
https://help.openai.com/en/articles/7730893
3. MIT Technology Review. (2025). GPT-5 raises new ethical questions for memory-enabled AI assistants.
https://www.technologyreview.com/2025/08/20/gpt5-memory-ethics