Editing OpenAI’s GPT-5 is here

[[file:OpenAI’s GPT-5 is here.jpg|500px]]

OpenAI unveiled GPT-5, its latest flagship AI model, on Thursday, powering the next evolution of ChatGPT. This “unified” model merges the rapid responses of the GPT series with the advanced reasoning capabilities of the o-series, marking a significant leap forward. GPT-5 positions ChatGPT as more than a conversational tool, steering OpenAI toward creating agent-like AI systems capable of independent task execution.

Unlike GPT-4, which excelled at delivering intelligent answers across diverse topics, GPT-5 empowers ChatGPT to perform complex tasks for users, such as developing software applications, managing calendars, or crafting research briefs. OpenAI has also streamlined the user experience with GPT-5, introducing a real-time router that automatically optimizes responses, balancing speed and depth without requiring users to adjust settings.

In a press briefing, OpenAI CEO Sam Altman declared GPT-5 “the best model in the world,” emphasizing its role as a “significant step” toward achieving artificial general intelligence (AGI) — AI capable of outperforming humans in most economically valuable tasks. “The existence of something like GPT-5 would have been unimaginable at any other point in history,” Altman stated, underscoring the model’s groundbreaking nature.

Launched on Thursday, GPT-5 is now the default model for all ChatGPT users, including those on the free tier, marking a historic shift. Nick Turley, OpenAI’s VP of ChatGPT, explained that this decision reflects the company’s commitment to its mission of democratizing advanced AI. “For the first time, we’re giving free users access to a reasoning model of this caliber,” Turley said, noting that earlier advanced models were restricted to paid subscribers. “This is a key part of living our mission to ensure AI benefits as many people as possible,” he added.

GPT-5 arrives as one of OpenAI’s most anticipated releases since ChatGPT’s explosive debut in 2022, which transformed the company into a household name. ChatGPT now boasts over 700 million weekly active users — roughly 10% of the global population — cementing its status as one of the world’s most widely used consumer products. The expectations for GPT-5 are immense, with its performance viewed as a critical indicator of AI’s broader progress.

The model’s reception could have far-reaching implications for Silicon Valley, Wall Street, and policymakers shaping AI regulation. Industry leaders, investors, and regulators are keenly observing whether GPT-5 delivers a transformative leap akin to GPT-4, which redefined expectations for what AI software can achieve. As a bellwether for the field, GPT-5’s capabilities could influence technological innovation, market dynamics, and the global conversation around AI’s future.

GPT-5 offers a slight edge on the competition

OpenAI’s GPT-5 demonstrates cutting-edge performance, narrowly surpassing competitors in key areas while trailing slightly in others, based on recent benchmarks. OpenAI highlights GPT-5’s exceptional coding capabilities, with CEO Sam Altman emphasizing its ability to generate entire software applications on demand, a process dubbed “vibe coding.”

On the SWE-bench Verified benchmark, which tests real-world coding tasks from GitHub, GPT-5 achieves a 74.9% success rate on its first attempt, marginally outperforming Anthropic’s Claude Opus 4.1 at 74.5% and significantly surpassing Google DeepMind’s Gemini 2.5 Pro at 59.6%. On the GPQA Diamond benchmark, featuring PhD-level science questions, GPT-5 Pro scores 89.4% on its first try, outpacing Claude Opus 4.1 (80.9%) and xAI’s Grok 4 Heavy (88.9%). Additionally, GPT-5 excels in health-related queries, with a hallucination rate of just 1.6% on the HealthBench Hard Hallucinations test, a substantial improvement over GPT-4o (12.9%) and o3 (15.8%).

However, GPT-5 Pro (with tools) scores 42% on Humanity’s Last Exam, a rigorous test spanning math, humanities, and natural sciences, falling slightly behind xAI’s Grok 4 Heavy at 44.4%. On Tau-bench, which measures agentic ability in simulated online tasks, GPT-5 shows mixed results: it scores 63.5% on airline website navigation, trailing OpenAI’s o3 (64.8%), and 81.1% on retail website navigation, slightly behind Claude Opus 4.1 (82.4%).

OpenAI’s focus on safety is notable, with GPT-5 exhibiting lower deception rates than other models, enhancing both safety and user trust, according to Alex Beutel, OpenAI’s safety research lead. While GPT-5’s performance is strong, its close competition with Claude Opus 4.1, Gemini 2.5 Pro, and Grok 4 Heavy suggests the AI landscape is highly competitive, with each model excelling in specific domains.

[[file:Humanity_last_exam.jpg|500px]]

OpenAI’s GPT-5 Pro scored 89.4% on its first try on the GPQA Diamond test, which includes tough science questions meant for PhD-level experts. It did better than Anthropic’s Claude Opus 4.1 (80.9%) and xAI’s Grok 4 Heavy (88.9%), showing it’s great at handling complex scientific topics.

For health-related questions, GPT-5 is highly accurate, making mistakes only 1.6% of the time on the HealthBench Hard Hallucinations test when it takes time to think. This is much better than OpenAI’s older models, GPT-4o (12.9%) and o3 (15.8%). While AI chatbots aren’t doctors, many people use them for health advice, so OpenAI made GPT-5 better at spotting health issues and helping users understand medical information.

GPT-5 also stands out in creative tasks like design and writing. Nick Turley, OpenAI’s VP of ChatGPT, said it feels more natural and has “better taste” than other AI models in these areas. “This model has really good vibes,” Turley added.

Overall, GPT-5 is a strong performer in science, health, and creativity, though other models like Claude Opus 4.1 and Grok 4 Heavy are close competitors in some areas.

GPT-5 shows marked improvement in accuracy over OpenAI’s earlier models, with a significant reduction in hallucinations — when AI generates incorrect or made-up information. OpenAI noted that its recent o-series models, like o3, had increasing hallucination issues, and the company wasn’t entirely sure why. However, GPT-5 (with thinking mode) has tackled this problem effectively, producing incorrect responses only 4.8% of the time in ChatGPT prompts. This is a big improvement compared to o3 (22%) and GPT-4o (20.6%).
On the Tau-bench test, which evaluates an AI’s ability to handle simulated online tasks, GPT-5’s performance is mixed. In navigating airline websites, GPT-5 scores 63.5%, slightly behind o3’s 64.8%. For retail website tasks, GPT-5 achieves 81.1%, just below Anthropic’s Claude Opus 4.1 at 82.4%. While GPT-5 excels in reducing errors and improving reliability, it remains closely competitive with other leading models in specific practical tasks.

OpenAI reports that GPT-5 is safer than its earlier models, showing a reduced tendency to deceive or act against users’ interests, a problem sometimes seen in AI reasoning models. Alex Beutel, OpenAI’s safety research lead, explained that lowering deception rates enhances both safety and user experience, making GPT-5 “more transparent and honest” in a way that builds trust.

Additionally, Beutel highlighted that GPT-5 is better at distinguishing between malicious users attempting to misuse ChatGPT and those making innocent requests. This allows GPT-5 to block more unsafe queries while rejecting fewer harmless ones, improving both security and usability compared to previous models.

Upgrades for consumers and developers

As part of the GPT-5 launch, ChatGPT is introducing user experience enhancements, including four new personality options: Cynic (sarcastic and witty), Robot (precise and emotionless), Listener (warm and empathetic), and Nerd (playful and knowledge-focused). These personalities, accessible via ChatGPT’s settings, automatically adjust the tone and style of responses to suit user preferences, eliminating the need for specific instructions.

Subscription tiers also see upgrades. ChatGPT Plus users ($20/month) enjoy higher usage limits for GPT-5 compared to free users, who face caps before switching to GPT-5 mini. Pro subscribers ($200/month) get unlimited GPT-5 access and the enhanced GPT-5 Pro, which leverages extra computational power for superior responses. Team, Edu, and Enterprise plan users will transition to GPT-5 as their default model starting next week, with tailored limits and features for organizational needs.

OpenAI is rolling out GPT-5 to its API in three variants gpt-5, gpt-5-mini, and gpt-5-nano — each designed to balance reasoning time and efficiency for different developer needs. A new API feature also allows developers to control response verbosity, tailoring whether GPT-5 provides concise or detailed outputs.
Pricing for the base GPT-5 model is set at $1.25 per million input tokens (equivalent to about 750,000 words, longer than the entire “Lord of the Rings” trilogy) and $10 per million output tokens, offering clarity for developers integrating the model into their applications.

The GPT-5 launch follows a hectic week for OpenAI, which also introduced gpt-oss, an open-weight reasoning model available for free download. Gpt-oss closely rivals the performance of OpenAI’s prior top models, o3 and o4-mini, and is cost-efficient for developers and enterprises. However, GPT-5 raises the bar in specific domains like coding, where it delivers frontier-level performance.
While GPT-5 matches other leading AI models in several benchmarks, its real-world impact remains to be seen. Benchmarks only provide a partial view, and developers’ practical applications will determine whether GPT-5 truly outshines competitors like Anthropic’s Claude Opus 4.1 or xAI’s Grok 4 Heavy.

Read the full article here: https://medium.com/@MsquareAutomation/openais-gpt-5-is-here-7451fb696eb9