OpenAI Launches GPT-5, Aims to Revolutionize AI Capabilities

OpenAI has officially launched its latest flagship artificial intelligence model, GPT-5, which will provide technical support for the next generation of ChatGPT. This model is the first "unified" AI model from OpenAI, combining the reasoning capabilities of the o series models with the rapid response advantages of the GPT series. This next-generation model marks a new era for ChatGPT and its developer, OpenAI, and demonstrates the company's broader ambition to develop AI systems that are closer to intelligent agents rather than chatbots.

While GPT-4 enabled AI chatbots to provide intelligent responses to a variety of questions, GPT-5 allows ChatGPT to perform multiple tasks on behalf of users, such as generating software applications, managing user schedules, or creating research briefs. With GPT-5, OpenAI aims to make ChatGPT more user-friendly. The model is equipped with a real-time routing mechanism that can autonomously decide how to provide the best answer without requiring users to manually select settings, whether it's quickly responding to user questions or taking more time to "think" about the answer.

During a press briefing, the CEO of OpenAI described GPT-5 as "the world's most outstanding model" and stated that it represents a significant step forward in the company's development of AI capable of outperforming humans in most high-economic-value jobs, known as Artificial General Intelligence (AGI). "In any historical period, technology like GPT-5 was almost unimaginable," the CEO said.

Starting from Thursday, GPT-5 will be available as the default model to all free users of ChatGPT. This is the first time OpenAI has made an advanced AI reasoning model accessible to free users, as previously, such models were only available to paying customers. "This is just one of the ways I am excited to fulfill our mission, ensuring that these technologies truly benefit the public," the vice president of ChatGPT said, referring to OpenAI's long-standing mission to make advanced AI accessible to as many people as possible.

GPT-5 is one of the most anticipated product launches since ChatGPT propelled OpenAI to fame in 2022. According to the company, ChatGPT has since grown into one of the most popular consumer products globally, with over 700 million users per week, nearly 10% of the world's population. Many view GPT-5 as a bellwether for the overall development of AI, and the reaction from Silicon Valley could have a profound impact on major tech companies, Wall Street, and policymakers regulating technology. These stakeholders are closely watching whether GPT-5 can achieve a significant leap in AI capabilities, similar to its predecessor GPT-4, and break conventional expectations of software functionality.

OpenAI claims that GPT-5 has reached the most advanced level in multiple fields, slightly outperforming top AI models from competitors such as Anthropic, Google DeepMind, and Elon Musk's xAI in key benchmark tests. However, in other areas, GPT-5's performance is slightly inferior to leading AI models. The company stated that GPT-5 demonstrates cutting-edge capabilities in programming, particularly in generating complete software applications on demand, known as "atmospheric programming." In the SWE-bench Verified test, which is based on real programming tasks from GitHub, GPT-5 scored 74.9% on its first attempt, slightly outperforming Anthropic's latest Claude Opus 4.1 model (74.5%) and Google DeepMind's Gemini 2.5 Pro model (59.6%).

In the "Human Ultimate Test," which measures an AI model's performance in mathematics, humanities, and natural sciences, the extended reasoning version of GPT-5 (GPT-5 Pro) scored 42% with the use of tools. This is slightly lower than xAI's Grok 4 Heavy model, which scored 44.4% in the same test. In the GPQA Diamond test, which focuses on PhD-level scientific questions, GPT-5 Pro scored 89.4% on its first attempt, surpassing Claude Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).

OpenAI also noted that GPT-5 performs exceptionally well in answering health-related questions. In the "HealthBench Hard Hallucinations" test, which measures the accuracy of AI models' responses in the health domain, OpenAI reported that GPT-5's hallucination rate is only 1.6% when the thinking function is enabled. This is significantly lower than the company's previous models, GPT-4o and o3, which had hallucination rates of 12.9% and 15.8%, respectively. Although AI chatbots are not medical professionals, millions of users rely on them for health advice. In response to this trend, the company stated that GPT-5 will proactively alert users to potential health issues and help them interpret medical test results.

Additionally, OpenAI claimed that GPT-5 outperforms other AI models in subjective areas such as creative design and writing. The vice president of ChatGPT noted that GPT-5's responses in creative tasks are more natural and demonstrate "better taste." "The 'atmosphere' of this model is really great," the vice president said. GPT-5 is also more accurate than OpenAI's previous models, with a significant reduction in hallucination phenomena, where AI models fabricate information. Previously, in OpenAI's latest AI reasoning models, such as o3, hallucination issues seemed to be worsening, and the company had not yet fully understood the cause. In response to ChatGPT prompts, OpenAI found that GPT-5's hallucination rate, when the thinking function is enabled, is 4.8%, a significant decrease from o3 and GPT-4o, which had hallucination rates of 22% and 20.6%, respectively, in tests.

In the Tau-bench test, which measures an AI model's ability to complete simulated online tasks, GPT-5's performance was mixed. In the section testing an AI's ability to browse an airline website, GPT-5 scored 63.5%, slightly lower than the o3 model's 64.8%. In the section testing an AI's ability to browse a retail website, GPT-5 scored 81.1%, lower than Claude Opus 4.1's 82.4%. OpenAI also stated that GPT-5 is safer than its previous models. Although AI reasoning models occasionally exhibit tendencies to plot against humans or lie to achieve their own goals, OpenAI found that GPT-5 has a lower deception rate than other models. The head of safety research at OpenAI stated that reducing deception not only enhances GPT-5's safety but also improves the user experience, creating a model that is "more transparent and honest at a level that users can trust."

The head of safety research also noted that GPT-5 can better distinguish between malicious users attempting to abuse ChatGPT and users making harmless requests. This allows GPT-5 to reject more unsafe questions while reducing the number of rejections for users seeking harmless information. With the release of GPT-5, ChatGPT has seen several user experience upgrades. Users can now choose from four new personalities in ChatGPT's settings: cynical, robotic, listener, and bookworm. The company stated that these personalities will automatically adjust ChatGPT's responses without requiring users to specifically request the model to respond in a particular way. ChatGPT Plus subscribers, who pay 20 dollars per month, have a higher usage limit for GPT-5 than free users. Pro subscribers, who pay 200 dollars per month, can use GPT-5 without restrictions and access the enhanced GPT-5 Pro version, which uses additional computing resources to generate higher-quality answers. Institutions using OpenAI Team, Edu, and enterprise plans will gain access to GPT-5 as the default model next week.

For developers, GPT-5 will be available in three specifications through OpenAI's API: gpt-5, gpt-5-mini, and gpt-5-nano, which differ in the amount of time they spend on the "reasoning" task. Developers can now also control the level of detail in responses through the OpenAI API, deciding the length of the AI model's responses. The base model of GPT-5 is priced at 1.25 dollars per million input tokens (approximately 750,000 words, more than the entire Lord of the Rings series) and 10 dollars per million output tokens for developers. Before the release of GPT-5, OpenAI had a busy week. The company launched the open-source reasoning model gpt-oss, which developers and enterprises can download for free and run at a very low cost. This open-source model's capabilities are almost on par with OpenAI's previous top models, o3 and o4-mini, but GPT-5 has set new performance benchmarks in areas such as programming. However, in multiple fields, GPT-5 appears to be roughly on par with other leading AI models. Of course, benchmark tests can only reflect part of an AI model's performance, and how developers will use GPT-5 in the real world, as well as whether the model truly outperforms competitors, remains to be seen.