OpenAI Launches GPT-5, Ushering in a New Era of AI Agents

 

OpenAI has unveiled GPT-5, its most advanced AI model to date, marking a major milestone in the evolution of ChatGPT and the broader AI landscape.

 

Released on Thursday, GPT-5 is OpenAI’s first unified model, combining the deep reasoning capabilities of its o-series with the fast response times of earlier GPT models. This new model represents a shift from traditional chatbots toward more capable AI agents, able to autonomously carry out complex tasks.

 

Unlike GPT-4, which primarily offered intelligent answers to questions, GPT-5 is built to act on users’ behalf — from writing full software applications and managing calendars to drafting research briefs. It’s also easier to use: a new real-time router automatically selects the best response method, whether that's instant answers or deeper reasoning.

 

“GPT-5 would have been unimaginable at any other point in history,” said OpenAI CEO Sam Altman, calling it “the best model in the world” and a key step toward artificial general intelligence (AGI).

 

Free Access and Mission Alignment

 

Starting immediately, GPT-5 becomes the default model for all free ChatGPT users — the first time OpenAI has provided a reasoning-capable model to non-paying users. Nick Turley, VP of ChatGPT, said this decision aligns with OpenAI’s mission to make advanced AI broadly accessible.

 

“This is about living the mission — making sure this technology benefits everyone,” said Turley.

 

Raising the Bar in AI Performance

 

GPT-5 is OpenAI’s most anticipated release since ChatGPT went viral in 2022. Now reaching 700 million weekly users — nearly 10% of the global population — the model is expected to set the tone for AI’s next phase.

 

OpenAI claims GPT-5 delivers top-tier performance in coding, health-related queries, and creative tasks. It slightly outperforms leading models like Anthropic’s Claude Opus 4.1, Google DeepMind’s Gemini 2.5 Pro, and xAI’s Grok 4 Heavy on many benchmarks — although it underperforms in certain areas.

 

Coding (SWE-bench Verified): GPT-5 scored 74.9%, ahead of Claude Opus 4.1 (74.5%) and Gemini 2.5 Pro (59.6%). Altman highlighted GPT-5’s ability to generate entire applications — what’s now being called “vibe coding.”

 

Health Advice Accuracy: On HealthBench Hard Hallucinations, GPT-5 (with reasoning) hallucinates just 1.6% of the time, far outperforming GPT-4o (12.9%) and o3 (15.8%).

 

PhD-Level Science (GPQA Diamond): GPT-5 Pro scored 89.4%, ahead of Claude Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).

 

Multidisciplinary Reasoning (Humanity’s Last Exam): GPT-5 Pro scored 42% using tools, slightly below Grok 4 Heavy’s 44.4%.

 

Less Hallucination, More Trust

 

One of GPT-5’s most notable improvements is its reduced tendency to "hallucinate" or generate incorrect information. While OpenAI’s previous o-series models had hallucination rates as high as 22%, GPT-5 (with reasoning) brings that down to 4.8%.

 

“GPT-5 is more transparent and honest,” said Alex Beutel, OpenAI’s safety research lead. He emphasized its improved ability to detect misuse while avoiding unnecessary rejections of harmless queries.

 

Agentic Ability: Mixed Results

 

On the Tau-bench, a benchmark for AI’s ability to perform online tasks:

 

Airline navigation: GPT-5 scored 63.5%, slightly below o3’s 64.8%.

 

Retail site navigation: GPT-5 scored 81.1%, just under Claude Opus 4.1’s 82.4%.

 

Despite these mixed results, GPT-5 performs well in creative fields. According to Turley, it shows more natural expression and "better taste" in writing and design.

 

“The vibes of this model are really good,” said Turley.

 

Upgrades for Users and Developers

 

Alongside GPT-5, ChatGPT is getting user-friendly upgrades:

 

New personalities (Cynic, Robot, Listener, Nerd) allow users to personalize responses without custom prompts.

 

Free users get basic access to GPT-5.

 

Plus ($20/month) subscribers receive higher usage limits.

 

Pro ($200/month) subscribers get unlimited access and exclusive features through GPT-5 Pro.

 

GPT-5 becomes the default for Team, Edu, and Enterprise plans next week.

 

For developers, GPT-5 will be available through OpenAI’s API in three sizes: gpt-5, gpt-5-mini, and gpt-5-nano, each with different reasoning depth. Developers can also now control verbosity — tailoring how brief or detailed responses should be.

 

Pricing: $1.25 per million input tokens and $10 per million output tokens.

 

A Busy Week at OpenAI

 

GPT-5’s debut follows OpenAI’s release of gpt-oss, an open-weight reasoning model available for free use. While gpt-oss approaches the capabilities of models like o3 and o4-mini, GPT-5 sets a new high bar in frontier AI performance — especially in software development and reasoning.

 

That said, benchmarks only tell part of the story. Real-world use will ultimately determine whether GPT-5 is a leap forward or simply a refinement in the race toward truly intelligent AI agents.