AI IQ Trends
9/28/2025 • 3 min read
It's easy to get lost in the numbers that companies use to report how good their models are. Is 65 a good score? Does that mean it's better than a human?
In this essay we aim to clarify the answer to the simple question "how intelligent actually is AI?" by explaining what the benchmark scores mean intuitively and showing the trends over time.
After taking a few minutes to read this essay, you will have an intuition for understanding this.
General Intelligence
OpenAI recently released a new test for AIs called "GDPval", to assess their ability to complete common work tasks. It evaluates their ability to produce realistic deliverables for different tasks across the 9 largest sectors that contribute to US GDP.
The latest results find that AI's ability to perform this kind of work is increasing steadily over time, and is nearing parity with human experts.
In this overall summary, AI performs slightly worse than a human expert on average across every area of economically valuable work. Below we will look specifically at which sectors and occupations AI outperforms and where humans win.
Sector Specific Intelligence
Here, we look at how often AI systems can produce equal or better work deliverables than a human expert in specific sectors.
Even if the AI system is not at parity with the human expert in your domain, its likely that there's a specialized tool out there that can accelerate their work, reduce costs, and improve outputs (thanks to fine-tuned models, AIs instructed with expert workflows etc).
Get in touch if you need help finding the right tool for the job.
Specialized Models Beat General Models
Despite the fact that general purpose AI models are not quite at parity with human expert work just yet, there are many specialized AI systems built by companies that excel in their domain or embed expert workflows
Software engineering is a good example of this. Despite not beating the human benchmark in the evaluation, from personal experience, I can ensure you that AI can write code as good as I can, because of the specialized models and workflows that are built into my AI editor.
Get in touch if you're curious about specific products from our partners that might map to your use case.
Other Key Types Of Intelligence
AI perform at and above human level on tasks that require web browsing (and this isn't even factoring in speed).
Conclusion
- AI systems are rapidly approaching human parity across the largest sectors of the economy.
- AI can use your computer, your browser, and your apps (via function calling).
Reach out to us if you have specific queries about how smart an AI is in a particular sector or occupation not explored here.