OpenAI Unveils GPT-5, Available to All: More Human-Like Expertise

CNBC AI News – In a highly anticipated livestream earlier today, OpenAI officially unveiled GPT-5, touting it as their most sophisticated model to date. The company claims GPT-5 boasts top-tier performance across a spectrum of critical domains, including programming, mathematics, writing, healthcare, and visual perception.

OpenAI CEO Sam Altman characterized GPT-5 as a “significant upgrade” over its predecessors, stating, “Engaging with it truly feels like conversing with an expert in any given field.”

Starting today, GPT-5 will be rolled out gradually to both free and paid users, with subscribers enjoying elevated usage limits.

Pro subscribers gain unlimited access to GPT-5, including the enhanced GPT-5 Pro. Plus users can set it as their default model for everyday inquiries, benefiting from a substantially higher usage allocation compared to free users.

The architecture of GPT-5 features a unified system comprised of an intelligent, high-performance model, a deep reasoning engine (“GPT-5 Thinking”), and a real-time router.

The intelligent, high-performance model is designed to handle the majority of queries, while the deep reasoning engine is reserved for tackling more intricate problems.

The real-time router intelligently determines which model to employ based on factors such as conversation type, complexity, tool requirements, and explicit user intent, further refining its performance through ongoing training and real-world signals.

Upon reaching usage caps, a streamlined version of each model will process remaining requests. OpenAI anticipates consolidating these functionalities into a single, seamless model in the future.

GPT-5 has demonstrated superior performance in benchmark tests compared to its predecessors. It delivers faster response times, more helpful answers to real-world questions, reduced instances of hallucination, improved instruction following, and minimized sycophancy.

Particularly noteworthy are the improvements observed in writing, programming, and healthcare consultations – three prevalent use cases for ChatGPT.

Programming: GPT-5 stands out as the most adept programming model to date, exhibiting substantial enhancements in complex front-end generation and large codebase debugging. It is capable of crafting aesthetically pleasing and highly responsive websites, applications, and games from single prompts.

Creative Expression & Writing: The model can translate rough ideas into richly resonant pieces with both literary depth and rhythmic flow. It displays an improved ability to manage writing tasks with ambiguous architectures, proving itself to be a powerful aide in everyday composition tasks.

Healthcare: GPT-5 excels in health-related inquiries, achieving scores on the HealthBench far surpassing those of previous models. It proactively identifies potential issues and requests clarification to provide more relevant and insightful answers, personalizing its responses for the user’s background, knowledge level, and geographic location. However, OpenAI clearly stipulates that it is not a replacement for the judgment of trained medical professionals.

The model has established new state-of-the-art results in areas such as mathematics (AIME 2025, 94.6% without tools), real-world programming (SWE-bench Verified 74.9%, Aider Polyglot 88%), multimodal understanding (MMMU 84.2%), and healthcare (HealthBench Hard 46.2%).

GPT-5 Pro, leveraging expanded reasoning capabilities, also established a new state-of-the-art in GPQA, achieving a score of 88.4% without tools.

GPT-5 Pro is purpose-built for the most demanding and complex tasks, superseding OpenAI’s o3-Pro offering. By allocating more “thinking time” and harnessing scaled, yet efficient, parallel computation for testing, it is designed to deliver the highest quality and most comprehensive responses.

GPT-5 Pro demonstrated top-tier performance across multiple challenging intellectual benchmarks. In an evaluation conducted over 1000 real-world reasoning prompts of economic value, external experts preferred GPT-5 Pro’s responses 67.8% of the time. The model exhibited a 22% reduction in key errors within critical areas like healthcare, science, mathematics, and programming.