For a long while, the launch of a new version,
GPT-5, has been anticipated, and today it has finally been rolled out. The team headed by Sam Altman, the founder of the cryptocurrency project Worldcoin, declared this as “our best AI system yet” at the launch. Let’s explore this new model, which is claimed to be far more capable than previous versions.
Advanced Capabilities of GPT-5
GPT-5 emerges as a model with a stronger reasoning ability for tackling more challenging problems. As of today, Plus subscribers can access GPT-5, while Pro subscribers gain access to the more advanced version, GPT-5 Pro.
The essence of advancing in artificial intelligence stems from the ability to process input in a deeper, more accurate manner, providing even more precise output. There is no end to progress in this area as the pathway to more profound, accurate, and realistic thinking remains open. At this point, GPT-5 represents the latest phase in this journey.
Enhanced Performance in Key Areas
Performance has significantly improved across three major use cases: writing, coding, and healthcare. To demonstrate the strength of improvements in design, the official website shared an application example prepared with a “single command.”
In healthcare, this model is set to be widely utilized, having achieved the best score on HealthBench compared to its predecessors. While GPT-5’s performance doesn’t greatly surpass the previous model o3, it has made notable progress towards perfection.
The distinction in coding performance is more pronounced compared to other models. The key detail here is the considerable improvement in performance through its “deep thinking” responses.
In fields like task monitoring and agent use, GPT-5’s sole differentiator remains its superior performance. The model’s most significant feature is its enhanced reasoning ability, outperforming other models across over 40 professions including law, logistics, sales, and engineering, in nearly half of the trials matching or exceeding expert-level performance.
A crucial update addresses the model’s tendency to “hallucinate.” GPT-5 has shown a significant reduction in misleading outputs, outperforming o3. While there is room for improvement, the rate of misleading responses during coding remains notably high compared to GPT-5’s normal operation.
The model now gives more accurate and serious responses, reducing nonsense in outputs. It no longer offers flattering, unhelpful responses, pulling back from behaviors making it seem artificial or unpleasant.
Lastly, Sam Altman noted that GPT-5 is the smartest model developed so far, focusing on practicality and mass accessibility. More intelligent models may be developed, but this one aims to benefit over a billion people, reflecting the real-world utility as the priority. Plus and Pro pricing remains unchanged, with Plus at $20 monthly, and Pro at $200.
