OpenAI officially launched GPT-5 on Thursday, August 7, making its most advanced AI model available to all ChatGPT users, including those on the free tier. The company describes it as its "smartest, fastest, most useful model yet" with built-in reasoning capabilities that CEO Sam Altman compared to having "PhD-level expert" intelligence.
The rollout began Thursday for Free, Plus, Pro and Team users, marking the first time free users have access to a reasoning model that performs internal chain-of-thought processing before responding. Enterprise and Education customers will receive access approximately one week later.
For developers, GPT-5 introduces significant architectural changes. The model family includes four variants: the main GPT-5 for complex reasoning tasks, GPT-5-mini for cost-sensitive applications, GPT-5-nano for ultra-low latency needs, and GPT-5-chat optimised for conversational interfaces. All three API versions—gpt-5, gpt-5-mini and gpt-5-nano—are designed for different cost and latency requirements.
Performance benchmarks show substantial improvements in areas critical to software development. GPT-5 achieved 74.9% accuracy on SWE-bench Verified coding tasks and 88% on Aider Polyglot, while setting new standards in mathematical reasoning with 94.6% on AIME 2025 without external tools. The model also demonstrated significant gains in multimodal understanding, scoring 84.2% on MMMU evaluations.
A key improvement addresses the persistent hallucination problem that has affected AI models. OpenAI claims GPT-5's responses are 45% less likely to contain factual errors than GPT-4o, and when using its reasoning mode, the error rate drops to 80% below the previous o3 model.
Altman demonstrated the model's coding capabilities during a press briefing, showing how it could generate "an entire piece of software" through what he termed "vibe coding"—creating complete applications from simple prompts. This capability aligns with broader industry trends, as Meta CEO Mark Zuckerberg has predicted that AI will write approximately half of his company's code by next year.
The launch comes as OpenAI faces intensifying competition from Anthropic's Claude, Google's Gemini, and other rivals, particularly among younger developers. The company reported expectations to reach 700 million weekly active ChatGPT users this week and is reportedly in discussions with investors about a stock valuation of roughly $500 billion.
Box CEO Aaron Levie, whose company conducts advanced AI testing, called GPT-5 a "complete breakthrough," noting the model's ability to "retain way more information" and apply "much higher level reasoning and logic capabilities" to complex documents and mathematical problems.
The unified model architecture eliminates the need for developers to manually select between different AI models for various tasks, potentially simplifying integration workflows and reducing the complexity of maintaining multiple model endpoints in production applications.