OpenAI’s GPT-5 is here

OpenAI has officially launched GPT-5, its new flagship AI model set to power the next generation of ChatGPT. Released on Thursday, GPT-5 is described as OpenAI’s first “unified” AI model, blending the advanced reasoning capabilities of its ‘o-series’ models with the rapid response times characteristic of its GPT series. This release signals a significant shift in OpenAI’s strategy, moving towards developing AI systems that function more like intelligent agents capable of performing complex tasks, rather than simple chatbots.

GPT-5 represents a substantial leap forward from its predecessor, GPT-4. While GPT-4 enabled AI chatbots to provide intelligent answers across a broad spectrum of queries, GPT-5 empowers ChatGPT to execute a wide array of tasks on behalf of users. These capabilities include generating software applications, managing a user’s calendar, and compiling detailed research briefs. OpenAI has also focused on enhancing user experience, equipping GPT-5 with a real-time router that intelligently determines the optimal approach for providing the best answer, whether that requires speed or more deliberate “thinking” time.

During a briefing with reporters, OpenAI CEO Sam Altman proclaimed GPT-5 as “the best model in the world,” characterizing it as a “significant step” towards achieving artificial general intelligence (AGI), the company’s ultimate goal of developing AI that can outperform humans in most economically valuable tasks. Altman stated, “Having something like GPT-5 would be pretty much unimaginable at any previous time in history.”

Starting Thursday, GPT-5 is available to all free users of ChatGPT as their default model. OpenAI’s VP of ChatGPT, Nick Turley, highlighted this move as a key effort to provide free users with access to an AI reasoning model for the first time, noting that such advanced capabilities were previously exclusive to paid tiers. Turley commented on the decision, “This is just one of the ways that I’m excited to live the mission, making sure that this stuff actually benefits people,” referencing OpenAI’s core mission to distribute advanced AI widely.

Expectations for GPT-5 are exceptionally high, marking one of OpenAI’s most anticipated product launches since ChatGPT gained widespread recognition in 2022. ChatGPT has since evolved into one of the world’s most popular consumer products, reaching over 700 million weekly users, which is nearly 10% of the global population. Many observers view GPT-5 as a critical indicator of broader AI progress, with its reception expected to influence major tech companies, financial markets, and regulatory bodies.

GPT-5 offers a slight edge on the competition

OpenAI asserts that GPT-5 achieves state-of-the-art performance across several domains, demonstrating a slight advantage over leading AI models from competitors like Anthropic, Google DeepMind, and Elon Musk’s xAI on key benchmarks. Specifically, GPT-5 exhibits frontier-level performance in coding, with CEO Sam Altman noting its particular strength in “vibe coding” – the ability to generate entire software applications on demand.

On the SWE-bench Verified benchmark, which tests real-world coding tasks sourced from GitHub, GPT-5 achieved a 74.9% success rate on its first attempt. This score marginally surpasses Anthropic’s Claude Opus 4.1 model, which scored 74.5%, and significantly outperforms Google DeepMind’s Gemini 2.5 Pro at 59.6%.

In broader academic and scientific assessments, GPT-5 shows varied results. On Humanity’s Last Exam, a challenging test of AI performance in mathematics, humanities, and natural sciences, a version of GPT-5 with extended reasoning (GPT-5 Pro) achieved a score of 42% when utilizing tools. This is slightly lower than xAI’s Grok 4 Heavy, which scored 44.4% on the same test. However, on the GPQA Diamond benchmark, designed to test AI models on PhD-level science questions, GPT-5 Pro demonstrated superior performance, scoring 89.4% on its first try, exceeding Claude Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).

OpenAI also highlighted GPT-5’s enhanced accuracy in healthcare-related queries. On the HealthBench Hard Hallucinations test, which measures the accuracy of AI responses concerning healthcare topics, GPT-5 (with thinking enabled) exhibited a hallucination rate of just 1.6%. This is a dramatic improvement compared to OpenAI’s previous models, GPT-4o (12.9%) and o3 (15.8%). Given that millions use AI for health advice, OpenAI states GPT-5 is more proactive in identifying potential health concerns and assisting users in understanding medical results.

Furthermore, OpenAI claims GPT-5 excels in more subjective domains like creative design and writing, exhibiting “better taste” and more natural responses than competing models. Nick Turley commented, “The vibes of this model are really good,” referring to its creative output.

Regarding the persistent issue of AI hallucinations, GPT-5 shows significant improvement. OpenAI reported that GPT-5 (with thinking) hallucinates and provides incorrect information only 4.8% of the time in response to ChatGPT prompts. This is a substantial reduction from the 22% rate of o3 and 20.6% rate of GPT-4o on the same test. This addresses a previous concern where OpenAI observed increasing hallucination rates in its reasoning models.

In terms of agentic capabilities, tested via the Tau-bench benchmark which simulates online task completion, GPT-5 presents mixed performance. On a segment involving navigation of an airline website, GPT-5 scored 63.5%, slightly below o3’s 64.8%. Similarly, on navigating retail websites, GPT-5 scored 81.1%, trailing Claude Opus 4.1’s 82.4%. OpenAI also reports that GPT-5 is safer than its predecessors, exhibiting lower rates of deception and a better ability to distinguish between malicious users and those with harmless requests, leading to more accurate refusal of unsafe prompts and fewer rejections of legitimate queries.

Upgrades for consumers and developers

The GPT-5 launch introduces several user experience enhancements for ChatGPT. Users can now select from four new personalities – Cynic, Robot, Listener, and Nerd – which dynamically adapt ChatGPT’s responses without explicit user prompts. Subscribers to ChatGPT Plus ($20/month) receive higher usage limits for GPT-5, while Pro subscribers ($200/month) gain unlimited access, including to an enhanced version, GPT-5 Pro, which leverages additional computational resources for superior outputs.

For developers, GPT-5 is accessible via OpenAI’s API in three variants: gpt-5, gpt-5-mini, and gpt-5-nano, each offering different “reasoning” depths. Developers can also now control the verbosity of AI responses through the API. The base GPT-5 model is priced at $1.25 per million input tokens and $10 per million output tokens.

This launch follows OpenAI’s recent release of gpt-oss, an open-weight reasoning model that offers capabilities nearly matching previous top models like o3 and o4-mini at a lower cost. While gpt-oss benefits developers seeking cost-effective solutions, GPT-5 establishes a new benchmark for frontier performance, particularly in areas like coding. However, GPT-5’s overall performance across various benchmarks suggests it is largely on par with other frontier AI models. The true impact and differentiation of GPT-5 will become clearer as developers integrate it into real-world applications.