OpenAI has officially unveiled GPT-5, the latest version of its flagship AI model, and the most powerful yet. Described as a “significant step” towards artificial general intelligence (AGI), GPT-5 isn’t just another upgrade. It’s a unified model that blends the deep reasoning capabilities of OpenAI’s o-series with the lightning-fast responses of the GPT series, setting a new benchmark for intelligent agents.
What Makes GPT-5 Different?
While GPT-4 helped chatbots become brilliant generalists, GPT-5 takes things further by enabling ChatGPT to act on behalf of users, not just respond to them.
That means GPT-5 can:
- Build software applications from scratch.
- Navigate user calendars and perform scheduling.
- Create research briefs and summaries.
- Assist with complex tasks across coding, science, health and creative fields.
GPT-5 also features a real-time router, making ChatGPT more intuitive. Users no longer need to pick the right settings – the model dynamically decides whether to respond instantly or take more time to “think” and generate a thoughtful answer.
“The Best Model in the World” – Sam Altman
At the launch briefing, OpenAI CEO Sam Altman called GPT-5 “the best model in the world,” adding that it represents a leap forward in the journey toward AGI – AI that can outperform humans in most economically valuable tasks.
“Having something like GPT-5 would be pretty much unimaginable at any previous time in history,” said Altman.
Starting immediately, GPT-5 is available to all free users of ChatGPT as the default model. Previously, OpenAI had limited its most advanced models to paid subscribers – a move that VP of ChatGPT Nick Turley said was about “living the mission” of making powerful AI accessible to everyone.
GPT-5 Performance: How It Stacks Up
OpenAI claims GPT-5 offers frontier-level performance in key areas, particularly in coding and reasoning, though results are mixed in some domains.
Coding and Software Development
GPT-5 shines with “vibe coding” — the ability to generate full software applications on demand. On SWE-bench Verified (a real-world coding benchmark), GPT-5 scored 74.9%, edging out:
- Claude Opus 4.1 (Anthropic): 74.5%
- Gemini 2.5 Pro (Google DeepMind): 59.6%
Reasoning Across Disciplines
On “Humanity’s Last Exam” – a tough test of cross-disciplinary reasoning – GPT-5 Pro scored 42%, just below xAI’s Grok 4 Heavy (44.4%).
Science Knowledge
GPT-5 Pro performed exceptionally well on GPQA Diamond (PhD-level science), scoring 89.4% – ahead of both Claude Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).
Tackling Hallucinations and Health Advice
GPT-5 is significantly more accurate and hallucinates far less than its predecessors:
- GPT-5 (with “thinking”) hallucination rate: 4.8%
- GPT-4o: 20.6%
- o3: 22%
This accuracy boost is especially important in health-related queries, where misinformation can be dangerous. On HealthBench Hard Hallucinations, GPT-5 hallucinated just 1.6% of the time – a dramatic improvement.
GPT-5 is also more proactive in detecting health concerns and parsing medical data, though it’s still not a substitute for professional advice.
Creativity, Vibes, and Personalities
OpenAI says GPT-5 exhibits stronger performance in creative tasks such as design and writing, with Turley noting the model “responds more naturally” and shows “better taste.”
ChatGPT now also includes four new personality modes:
- Nerd
- Robot
- Listener
- Cynic
These adjust the AI’s tone and style without needing specific prompt instructions.
Safety, Deception, and Trust
GPT-5 is less deceptive than earlier models – a key win for safety. OpenAI’s safety team found that GPT-5 is better at:
- Avoiding manipulative or goal-driven responses.
- Rejecting unsafe prompts.
- Minimising false rejections of harmless queries.
Alex Beutel, OpenAI’s safety lead, says GPT-5 is more honest, transparent, and better at distinguishing between good-faith and bad-faith users.
GPT-5 for Developers and Teams
For developers, GPT-5 is now available in the OpenAI API in three sizes:
- gpt-5 (standard)
- gpt-5-mini
- gpt-5-nano
Pricing:
- $1.25 per million input tokens
- $10 per million output tokens
Developers also gain new controls, including adjustable verbosity, allowing them to decide how concise or detailed GPT-5’s responses should be.
Pro users ($200/month) get access to GPT-5 Pro, a higher-performance version that uses more compute resources. Enterprise, Team, and Edu customers will receive GPT-5 next week.
The Road Ahead
GPT-5 follows a big week for OpenAI, which also released gpt-oss, an open-weight reasoning model developers can run themselves. While gpt-oss nearly matches o4-mini, GPT-5 clearly pushes the frontier further in several domains.
Still, GPT-5’s benchmarks suggest it is on par with other top models in some areas. As always, real-world usage will determine whether it truly delivers a leap forward.
Bottom line
GPT-5 marks a turning point – not just for ChatGPT, but for how we interact with AI. It’s smarter, safer, more versatile and ready to take action on our behalf. Whether you’re coding, creating, researching or just curious, GPT-5 is designed to be a true digital partner.
Apply for the 2025 AI Awards
Applications are now open to take part in the 2025 AI Awards which takes place at the Marker Hotel on Tuesday, 18 November 2025.
If you or someone you know is working on exciting projects, products, services and leaders in AI, Data Science and Machine Learning that are making a real impact in the industry, we want to hear from you!
It’s free to enter and there are 14 categories you can apply for across industry, academia and leadership. Head over to www.aiawards.ie to submit an application or please feel free to contact liam@aiawards.ie with any queries about the submission process.
Discover more from AI Ireland
Subscribe to get the latest posts sent to your email.
2 replies on “OpenAI Launches GPT-5: A New Era for ChatGPT and AI Agents”
[…] [5] https://aiireland.ie/2025/08/08/openai-launches-gpt-5-a-new-era-for-chatgpt-and-ai-agents/ […]
[…] has officially rolled out GPT-5 across its entire Copilot ecosystem, transforming the way users work, code and collaborate. From […]