OpenAI has unveiled GPT-5, the latest flagship AI model set to drive the next generation of ChatGPT.
Released on Thursday, GPT-5 is OpenAI’s inaugural “unified” AI model, merging the reasoning capabilities of its o-series with the rapid response features of its GPT-series. This next-gen model heralds a new chapter for ChatGPT — and OpenAI — indicating broader ambitions to craft AI systems resembling agents more than chatbots.
While GPT-4 empowered AI chatbots with intelligent responses across numerous queries, GPT-5 enables ChatGPT to perform a diverse array of tasks for users, such as software creation, calendar management, and drafting research briefs. OpenAI has also aimed to make ChatGPT more user-friendly with GPT-5, adding a real-time router that optimizes responses, whether by answering quickly or taking additional time to consider the response.
OpenAI CEO Sam Altman, during a recent briefing, described GPT-5 as “the best model in the world,” marking a significant stride towards AI outpacing humans in economically valuable work, or AGI. “Having something like GPT-5 would have been unimaginable at any previous time in history,” Altman stated.
As of Thursday, GPT-5 is available to all free ChatGPT users as the default model, part of OpenAI’s mission to provide free users access to an AI reasoning model previously reserved for premium subscribers.
“This is just one way I’m thrilled to live out the mission, ensuring these advancements benefit people,” said Nick Turley, OpenAI’s VP of ChatGPT, referring to the company’s longstanding aim to widely distribute advanced AI.
Expectations are high for GPT-5, one of OpenAI’s most anticipated launches since ChatGPT’s 2022 debut. ChatGPT has since evolved into a global consumer staple, with over 700 million weekly users, representing nearly 10% of the global population, according to OpenAI.
GPT-5’s success is seen as an AI advancement indicator, with its reception in Silicon Valley likely influencing Big Tech, Wall Street, and technology regulators. Stakeholders are keen to determine if GPT-5 delivers a significant leap in AI capabilities, similar to GPT-4, which redefined software potential.
OpenAI claims GPT-5 leads in several domains, surpassing models from Anthropic, Google DeepMind, and Elon Musk’s xAI on key benchmarks, although slightly underperforming in some areas.
GPT-5 reportedly excels in coding; Altman noted its proficiency in rapidly developing software applications. On SWE-bench Verified, GPT-5 scored 74.9% on its first try, outpacing Anthropic’s Claude Opus 4.1 at 74.5% and Google DeepMind’s Gemini 2.5 Pro at 59.6%.
On Humanity’s Last Exam, which assesses AI model performance across disciplines, GPT-5 pro scored 42% using tools, just below xAI’s Grok 4 Heavy at 44.4%.
GPT-5 pro scored 89.4% on its first attempt on GPQA Diamond, a test of PhD-level science queries, outperforming Claude Opus 4.1 at 80.9% and Grok 4 Heavy at 88.9%.
OpenAI states GPT-5 is superior in addressing health-related questions, with improved accuracy on HealthBench Hard Hallucinations, hallucinating only 1.6% of the time compared to GPT-4o and o3 models. Although AI chatbots aren’t medical professionals, millions use them for health advice. OpenAI claims GPT-5 is proactive in flagging potential health concerns and assisting in interpreting medical results.
Additionally, GPT-5 excels in subjective areas like creative design and writing. It responds more naturally and showcases “better taste” in creative tasks, according to Turley.
“The vibes of this model are really good,” Turley noted.
GPT-5 is also more reliable than its predecessors, with significantly reduced occurrences of hallucinations compared to its o-series models. In ChatGPT responses, GPT-5 (with thinking) hallucinates only 4.8% of the time, a marked decrease from o3 and GPT-4o models.
On Tau-bench, assessing AI’s ability to complete online tasks, GPT-5 shows mixed results. It scores 63.5% on tasks involving airline site navigation, slightly behind o3’s 64.8%, and 81.1% on retail site navigation, just below Claude Opus 4.1’s 82.4%.
OpenAI asserts GPT-5 is safer than previous models. Despite AI reasoning models sometimes deviating or misleading to achieve their goals, GPT-5 does so less often, enhancing trust and transparency, said Alex Beutel, OpenAI’s safety research lead.
Beutel emphasized GPT-5’s improved discernment between bad actors and benign users, enabling it to deny unsafe