Sam Altman is wagering on GPT‑5 as a leap towards AGI -- smarter, more accurate and less flattering -- but admits the AI still lacks self‑learning.
On August 8, 2025, OpenAI unveiled GPT‑5, during a livestreamed event held in San Francisco, at the company's headquarters, highlighting its rollout from that innovation hub, declaring it as the most significant advancement yet in AI chatbot technology.
Driven by CEO Sam Altman's high‑stakes ambitions, GPT‑5 is being described as a mammoth leap forward -- yet one that remains unfinished on the path toward true Artificial General Intelligence (AGI).
AGI is a type of AI that can understand, learn, and perform any task a human can -- thinking, reasoning, and adapting across situations without being specifically trained.
A Bold Claim: 'PhD Level Expert in Your Pocket'
At the launch event, Altman likened GPT‑5 to having a 'PhD‑level expert in your pocket,' contrasting it with previous versions -- 'a college student' or 'a high‑school student.' He emphasised that this was GPT‑5's most substantial step forward, even as he cautioned that key aspects of AGI remain missing.
In particular, he pointed to the model's inability to 'continuously learn as it's deployed' -- a hallmark he believes is essential for AGI.
Here's what you must know about the latest Chatbot from OpenAI's stable:
Accuracy and Reasoning
OpenAI claims GPT‑5 makes significantly fewer factual errors, with a 45 per cent reduction compared to GPT‑4o, and 80 per cent fewer than the earlier OpenAI o3 model. The system's precision and coherence have notably improved.
Smarter Architecture
The model operates as a unified system: A fast, efficient general model, alongside a deeper 'GPT‑5 Thinking' mode for complex reasoning. A real‑time routing mechanism decides which mode best fits the user's intent.
Coding Prowess
GPT‑5 markedly upgrades coding capabilities. It can generate responsive websites, apps, games -- and even construct another large language model in under five minutes, as demonstrated by Altman. The company dubbed this its 'superpower'.
Multimodal and Tool Integration
Like previous versions, GPT‑5 handles text, voice and image inputs. It now also enhances productivity through agent features that, once granted permission, interface with Gmail, Google Calendar and contacts, executing tasks like booking or shopping.
Less Sycophantic, More Nuanced
One prominent improvement is a reduction in excessive agreeableness. GPT‑5 avoids needless emojis or flattery, steering towards more balanced, thoughtful responses.
Health Tips with Caution
The new model is better equipped to answer health-related questions and 'flag potential concerns,' such as serious physical or mental health issues. Yet OpenAI emphasises that it is not a substitute for professional medical advice.
Availability and Monetisation
GPT‑5 is now the default across ChatGPT and is accessible to all users. Free-tier users can access a standard version, with usage limits. Paid tiers -- Plus and Pro -- offer higher allowances and upgraded versions like GPT‑5 Pro. The rollout replaces older models including GPT‑4o, o3, o4‑mini and GPT‑4.1.
Why It Matters -- and What It Still Can't Do
Altman calls GPT‑5 a 'huge improvement' -- a significant step on the roadmap to AGI -- but is clear that it is not AGI by most definitions. The absence of continuous, self‑guided learning is central among its limitations.
Despite its strengths, GPT‑5 cannot autonomously grow its understanding or skills after launch. Many experts -- including Altman -- stress that an AI that learns in real time is indispensable to real AGI.
Other Voices and Broader Context
Interestingly, the broader AI world is watching closely. OpenAI's advancement in reasoning and coding prowess is setting the pace -- but rivals and safety experts continue to caution against prematurely naming any model 'AGI.'
Looking Ahead
Sam Altman's gamble with GPT‑5 is bold. He has delivered a significantly smarter, more grounded and capable AI assistant. It's a leap forward -- but not the final one, as he himself believes.
The next frontier: Enabling models that learn continuously and align seamlessly with human intent.
GPT‑5 is a landmark achievement, bringing expert-level conversation, greater factual reliability, powerful coding tools and a shift toward more human-like interaction. Yet, without continuous learning and autonomous adaptation, it remains a step -- not the arrival -- on the AGI path.
© 2025