Last Thursday, OpenAI launched the latest version of its hyper-popular AI chatbot, ChatGPT. Sam Altman, OpenAI’s CEO, made some pretty big claims about GPT-5 during its unveiling.
“GPT-3 was sort of like talking to a high school student: There were flashes of brilliance, lots of annoyance, but people started to use it and get some value out of it,” he said. “With GPT-4, maybe it was like talking to a college student: real intelligence, real utility. But with GPT-5, now it’s like talking to an expert: a legitimate Ph.D-level expert in anything, any area you need, on demand that can help you with whatever your goals are.”
While OpenAI insists GPT-5 is the company’s best AI yet, with more expertise, more capability, and fewer hallucinations and mistakes compared to past models, users aren’t so convinced. Many of them were miffed that, upon release of GPT-5, OpenAI hid its other models, including its previous mainstream AI chatbot, GPT-4o. (It has since walked back this change.) But some ChatGPT users have also complained about GPT-5’s “cold” or “blunt” conversational style compared to earlier generations, with threads comparing it to an “overworked secretary.” Others have also pointed out the model’s stricter safety and refusal policies, despite improvements in transparency, and that other rivals including Anthropic’s Claude, Google’s Gemini, or Meta’s Llama are still better for certain workflows around live coding or customizable deployments.
If you count yourself as one of the masses underwhelmed with the advent of GPT-5, not to worry: We’ve highlighted several advantages GPT-5 has over other AI models; conversely, we’ve also outlined where other models shine in comparison to OpenAI’s latest and greatest.
What ChatGPT-5 can do that older models couldn’t
1. Unified, adaptive reasoning
In its GPT-5 unveiling, OpenAI emphasized how previous versions of ChatGPT required users to select distinct models for different tasks; one might choose “o3” for responses that demanded more research, time be damned, whereas “4o” could yield smart answers with very little downtime. The headline innovation with GPT-5, according to OpenAI, is that it claims to eliminate the need to choose one model over the other; OpenAI says GPT-5 is the best all-around option for both quick responses and deep thinking. OpenAI claims GPT-5 is particularly good at coding and math.
2. More accurate answers with fewer hallucinations
OpenAI says GPT-5 excels at extended reasoning, with dramatically reduced hallucination rates and improved accuracy. For example, HealthBench scores show up to 80% fewer factual errors in complex scenarios compared to earlier models, particularly GPT-o3, which was a model that would take a long time to respond but typically offered precise answers and few errors.
3. Rich personalization
GPT-5 introduces customizable personalities (Cynic, Robot, Listener, Nerd) and enhanced voice features, letting users tailor the chatbot’s tone, style, and conversational approach without resorting to prompt engineering. The user interface itself can now adapt color schemes and conversation settings. The voice model, according to OpenAI, sounds “incredibly natural” and can dynamically adjust speaking speed for language learning. Memory enhancements allow ChatGPT to learn about users over time and understand their goals. For paid users, it will soon have access to Gmail and Google Calendar, enabling it to plan schedules, integrate appointments, and manage emails, making it significantly more “useful and personal”.
4. Workflow Integration
GPT-5 supports direct integration with apps like Gmail and Google Calendar, which allows ChatGPT to access and manage schedules or emails on command. Businesses can also customize GPT-5 with proprietary data, making the AI feel more like a personalized digital assistant rather than a generic chatbot. In terms of doing actual work, OpenAI calls GPT-5 “the best coding model on the market today” and claims it can write an “entire computer program from scratch.” It can work autonomously for extended periods, and a “self-improvement loop” lets GPT-5 build projects while sending errors back to itself to iterate on its own code.
What other AI models can do that GPT-5 cannot
Despite its power and versatility, GPT-5 isn’t the only game in town:
Claude Opus 4.1, from Anthropic, is widely regarded to be the superior model for sustained autonomous coding and agentic workflows. Enterprise features like “Claude Code” support integrated code review and real-time security, offering specialized developer tools that GPT-5 lacks, at least right now. Claude also projects its “artifacts”—code visualizations—live within its integrated development environment (IDE), improving clarity for engineering teams, and Claude maintains artifacts locally so it can follow long-running tasks across sessions, which may surpass GPT-5’s current workflow persistence in some settings.
Google’s Gemini 2.5 Pro handles seamless video analysis with ease, which is still an emerging area for GPT-5, where its capabilities are still more limited. Gemini’s multimodal engine natively handles short video clips, diagrams, and complex visual reasoning, and demonstrates deep contextual understanding. It’s also great at getting real-time data from the internet, thanks to its ties to Google’s Search product. GPT-5, while brilliant, typically doesn’t do so well with breaking news or minute-by-minute market data.
Llama 3, from Meta, could be best for research environments and organizations with strict privacy needs since, unlike GPT-5’s closed API-driven approach, Llama is an open-source model, which offers developers full control over customization, security, and deployment. Llama 3 is also excellent at handling lengthy documents and multi-turn conversations, so users can deploy it on their own infrastructure. Llama being open-source also allows for faster experimentation and improvements.
Despite all the improvements OpenAI touts for ChatGPT-5, the AI arms race is still very much in full force. Anthropic, Google, and Meta are all keeping pace, especially when it comes to customization, coding, and breaking news. Of course, as with all things AI-related, it’s best to experiment with different models to find the right model, and workflow, that works best for you.
For this story, Fortune used generative AI to help with an initial draft. An editor verified the accuracy of the information before publishing.
