Three years after ChatGPT: what’s in the modern AI toolkit? – The Media Online
The AI models of 2025 can do things we only dreamed about three years ago/Unsplash
It’s been just three years since ChatGPT launched, but it feels like a lifetime. The tech has exploded, not just with new models and capabilities, but with entirely new ways of working, creating and thinking.
In September 2022, Queen Elizabeth had just died, the war in Ukraine was escalating and the Johnny Depp vs Amber Heard trial dominated headlines. But generative AI, something that now feels so normal, hadn’t even officially launched yet. That would only happen on 30 November 2022.
If someone had shown you a tool on your laptop back then that could instantly draft documents, answer complex questions, generate images and even write code, would you have believed it?
The AI models of 2025 can do things we only dreamed about three years ago. But to understand how these models have evolved, I like to look at the AI benchmarks – the standardised tests used to compare the performance of AI models across capabilities such as reasoning, language understanding and coding.
Three years ago, ChatGPT was famously bad at maths. Today, the model and others, such as Claude, Gemini and Grok 4, are acing graduate-level problems and even producing high-quality visual code outputs such as vector illustrations.
One particularly tough benchmark is called ‘Humanity’s Last Exam’ – a set of 2 500 of the world’s hardest questions. When the models were first tested, they barely scraped 5% accuracy. Today, however, the current top model, Grok 4, is hitting 25%. It may not yet be a passing grade, but it is a major leap forward for such challenging material.
It’s no longer a case of choosing one tool and sticking with it. Different models have different strengths and at Flow Communications we use a mix, depending on what we’re trying to achieve:
ChatGPT and Gemini: our go-to generalists for fast answers, idea generation and content support
Claude: currently my favourite for writing and reviewing code. It’s brilliant at software development tasks
11 Labs: the best I’ve seen for generating natural, lifelike voice-overs for audio content
Gemini (image and video): leading the pack when it comes to producing high-quality, realistic visuals
DeepSeek: an impressive open-source option that’s competing and quickly closing the gap with the top proprietary models
Early AI models competed with each other in terms of brute force (more computer power and more data). Now, it’s about how the model thinks. The best tools can break problems into steps, reason through them and even run small programs to get results.
Whether it’s browsing the web, writing code or simulating actions, modern AI models have to use other tools on your computer, like an assistant that doesn’t just know things, but can do things.
We’ve gone from feeding AI plain text to giving it images, audio and video. We now expect an AI platform to respond across formats and generate content in multiple forms. You must be able to upload a photo or ask for a video and get it instantly.
The models that once required a data centre to run now work on personal laptops. AI is now going to show up everywhere: in your phone, fridge or car. The lower the power requirement, the more embedded AI will become in everyday life.
What used to be the playground of a few big tech companies is now open to all. Open-source models from China and elsewhere are rivalling, and in some cases surpassing, their proprietary counterparts. Innovation is accelerating because more people can contribute.
If 2025 has been the year of integration and experimentation, I believe 2026 will be the year of agentic AI. These are AI systems that act on your behalf – helping you buy groceries, run errands online or even handling your admin or banking.
More people will build their own software using natural language prompts without having to know coding. And because the tech is so much more efficient, AI will soon be part of every connected device in your home.
The most valuable skill is knowing how and when to use the tools. That’s what we focus on when hiring, and it’s what I encourage our clients to think about too.
Richard Frank is the chief technology officer of Flow Communications (www.flowsa.com), one of South Africa’s leading marketing and communications agencies.
Richard Frank is the Chief Technology Officer of Flow Communications (www.flowsa.com), one of South Africa’s leading independent agencies.
The Media Online is the definitive online point of reference for South Africa’s media industry offering relevant, focused and topical news on the media sector. We deliver up-to-date industry insights, guest columns, case studies, content from local and global contributors, news, views and interviews on a daily basis as well as providing an online home for The Media magazine’s content, which is posted on a monthly basis.
Editor: Glenda Nevill
glenda.nevill@cybersmart.co.za
Sales and Advertising:
Tarin-Lee Watts
wattst@arena.africa
Download our rate card
TimesLIVE
Sunday Times
SowetanLIVE
BusinessLIVE
Business Day
Financial Mail
HeraldLIVE
DispatchLIVE
Wanted Online
SA Home Owner
Business Media MAGS
Arena Events
Copyright © 2015 – 2023 The Media Online. All rights reserved. Part of Arena Holdings (Pty) Ltd
Copyright © 2015 – 2023 The Media Online. All rights reserved. Part of Arena Holdings (Pty) Ltd
Login to your account below
Please enter your username or email address to reset your password.
