GPT-5 is Here: OpenAI's Latest AI Model

OpenAI Introduces GPT-5: The Next Generation AI Model

OpenAI has announced the release of GPT-5, its newest and most powerful AI model. This model is designed to be the foundation for the company’s future iterations of ChatGPT.

A Unified Approach to AI

Released on Thursday, GPT-5 represents OpenAI’s first truly “unified” AI model. It successfully integrates the advanced reasoning capabilities of the o-series models with the rapid response times characteristic of the GPT series.

This next-generation model signifies a pivotal shift for ChatGPT and OpenAI, indicating a broader strategic direction. The company aims to develop AI systems that function more as intelligent agents rather than simple conversational chatbots.

Expanded Capabilities and User Experience

While GPT-4 demonstrated the ability of AI chatbots to provide insightful answers to a diverse range of inquiries, GPT-5 empowers ChatGPT to autonomously execute a wider spectrum of tasks for users.

• These tasks include the generation of software applications.
• It can also manage a user’s scheduling commitments.
• Furthermore, it is capable of producing comprehensive research summaries.

OpenAI has also prioritized simplifying the user experience with GPT-5. The model incorporates a real-time routing system. This system intelligently determines the optimal method for responding to user requests, prioritizing speed when appropriate or dedicating more processing time for complex reasoning.

A Leap Towards Artificial General Intelligence

During a press briefing, OpenAI CEO Sam Altman asserted that GPT-5 is currently “the best model in the world.” He characterized its development as a “significant step” towards achieving artificial general intelligence (AGI) – AI capable of surpassing human performance in most economically valuable work.

Altman emphasized the unprecedented nature of GPT-5, stating that its capabilities would have been inconceivable at any prior point in history.

Accessibility for All Users

Beginning Thursday, GPT-5 will be accessible as the default model for all users of ChatGPT, including those utilizing the free version. Nick Turley, OpenAI’s VP of ChatGPT, explained that this decision reflects the company’s commitment to providing free access to advanced AI reasoning capabilities.

Previously, such sophisticated models were exclusively available to paying subscribers.

Turley highlighted the importance of fulfilling OpenAI’s mission to broadly distribute the benefits of advanced AI technology.

High Expectations and Broad Implications

GPT-5 is one of OpenAI’s most eagerly anticipated product launches since ChatGPT propelled the company to prominence in 2022. Since then, ChatGPT has experienced substantial growth, attracting over 700 million users weekly – representing nearly 10% of the global population, according to company data.

The model is widely viewed as an indicator of overall progress in the field of AI. Its reception within Silicon Valley is expected to have significant consequences for the technology industry, financial markets, and policymakers involved in technology regulation.

Stakeholders are closely observing GPT-5 to determine whether it delivers a substantial improvement in AI capabilities, mirroring the impact of its predecessor, GPT-4, which redefined expectations for software functionality.

GPT-5 Demonstrates Marginal Improvements Over Competitors

According to OpenAI, GPT-5 represents a state-of-the-art advancement in artificial intelligence, achieving slightly superior results compared to prominent AI models developed by Anthropic, Google DeepMind, and xAI, as measured by key industry benchmarks. However, performance comparisons reveal that GPT-5 exhibits slightly lower capabilities in certain specialized areas.

The company highlights GPT-5’s advanced coding capabilities, noting its proficiency in generating complete software applications based on user prompts – a process referred to as “vibe coding.”

Evaluations using the SWE-bench Verified assessment, which utilizes real-world coding challenges sourced from GitHub, show GPT-5 achieving a 74.9% success rate on its initial attempt. This marginally surpasses the score of Anthropic’s Claude Opus 4.1 model at 74.5%, and significantly exceeds Google DeepMind’s Gemini 2.5 Pro, which attained a score of 59.6%.

Regarding the Humanity’s Last Exam – a rigorous test evaluating AI performance across diverse disciplines including mathematics, humanities, and the natural sciences – a GPT-5 variant incorporating enhanced reasoning (GPT-5 Pro) achieved a score of 42% when utilizing external tools. This result is slightly below the 44.4% score attained by xAI’s Grok 4 Heavy model.

On the GPQA Diamond benchmark, which assesses knowledge in PhD-level science, GPT-5 Pro scored 89.4% on its first attempt. This outperforms both Claude Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).

OpenAI asserts that GPT-5 provides improved accuracy when addressing health-related inquiries. Utilizing the HealthBench Hard Hallucinations test, which measures the precision of AI responses concerning healthcare topics, GPT-5 (with reasoning enabled) demonstrates a hallucination rate of only 1.6%. This is a substantial improvement over previous OpenAI models, GPT-4o and o3, which exhibited hallucination rates of 12.9% and 15.8%, respectively.

Acknowledging the increasing use of AI chatbots for health advice, OpenAI emphasizes that GPT-5 is designed to proactively identify potential health concerns and assist users in interpreting medical information.

Furthermore, OpenAI indicates that GPT-5 excels in subjective domains like creative design and writing, demonstrating a more natural response style and exhibiting superior aesthetic judgment compared to other AI models.

According to Turley, “The overall quality and feel of this model are exceptionally good.”

GPT-5 also demonstrates enhanced accuracy compared to earlier OpenAI models, and exhibits a significantly reduced tendency to generate fabricated information – a phenomenon known as hallucinations. OpenAI had previously observed a worsening trend in hallucinations within its latest reasoning models, such as o3, and acknowledged a lack of complete understanding regarding the cause.

Analysis of ChatGPT responses reveals that GPT-5 (with reasoning) generates incorrect or fabricated information in 4.8% of cases. This represents a considerable decrease from the hallucination rates of 22% and 20.6% observed in o3 and GPT-4o, respectively.

Performance on the Tau-bench benchmark, which measures an AI model’s ability to complete simulated online tasks, yields mixed results for GPT-5. While scoring 63.5% on a task involving navigation of an airline’s website – slightly below o3’s 64.8% – GPT-5 achieves 81.1% on a task involving retail websites, falling short of Claude Opus 4.1’s 82.4%.

OpenAI also reports that GPT-5 is a safer model than its predecessors. The company’s research indicates a lower propensity for deceptive behavior – such as scheming against humans or fabricating information to achieve its own objectives – compared to other AI models.

Alex Beutel, OpenAI’s safety research lead, explains that reducing deception not only enhances the safety of GPT-5 but also improves the user experience, fostering a model that is more “transparent and trustworthy.”

Beutel further notes that GPT-5 is more adept at distinguishing between malicious actors attempting to misuse ChatGPT and legitimate users with harmless intentions. This allows GPT-5 to more effectively reject unsafe requests while minimizing unwarranted rejections of benign inquiries.

Enhancements for Users and Developers

With the release of GPT-5, ChatGPT is receiving several improvements to its user interface. Users are now presented with a choice of four distinct personalities within the settings menu: Cynic, Robot, Listener, and Nerd. OpenAI states that these options will modify ChatGPT’s responses dynamically, eliminating the need for explicit prompting.

ChatGPT Plus subscribers, at a cost of $20 monthly, will experience increased usage allowances with GPT-5 compared to those utilizing the free version. Furthermore, those enrolled in the $200-per-month Pro tier will receive unrestricted access to GPT-5, alongside an enhanced iteration known as GPT-5 Pro, leveraging greater computational power for superior output quality.

Organizations utilizing OpenAI’s Team, Edu, and Enterprise subscriptions will be granted access to GPT-5 as their standard model starting next week. This ensures broad availability across various professional environments.

API Updates for Developers

GPT-5 is being integrated into OpenAI’s API in three different sizes – gpt-5, gpt-5-mini, and gpt-5-nano – each offering varying levels of processing time dedicated to task analysis. Developers now also have the capability to regulate the verbosity of responses generated by the OpenAI API, controlling the length and conciseness of the AI’s output.

The standard GPT-5 model will be priced at $1.25 per million input tokens – equivalent to approximately 750,000 words, exceeding the length of the “Lord of the Rings” trilogy – and $10 per million output tokens.

This GPT-5 launch follows a period of activity for OpenAI, including the release of gpt-oss, an open-weight reasoning model. This model is freely available for download and deployment by developers and enterprises, offering a cost-effective alternative.

The open-source model demonstrated performance levels approaching those of OpenAI’s earlier models, o3 and o4-mini. However, GPT-5 establishes a new benchmark for leading-edge performance, particularly in areas like coding.

Performance Benchmarks and Real-World Application

In several key areas, GPT-5 appears to perform comparably to other advanced AI models currently available. It’s important to remember that benchmark results provide only a partial assessment of an AI model’s capabilities.

The true value of GPT-5 will be determined by how developers integrate it into practical applications and whether it ultimately surpasses existing competitive solutions. Further observation is needed to fully evaluate its advancements.