November 2, 2026 · SemiAnalysis / Unite.AI
SemiAnalysis Launches GDPx Benchmark Spanning 11 Agentic Domains
My take: When the best AI model across 11 specialized domains tops out at 12%, that tells us something important: AI is still a support tool, not an autonomous expert. In finance and law, every single one of the 13 models tested scored zero on the most complex tasks. That doesn't mean AI is useless in those fields — it means it works for what it was designed for: helping you work more efficiently, not replacing specialized judgment.
For those using AI in their daily work, this benchmark is a practical guide: in chip design, quantitative trading, and software engineering, models are already performing well. In complex legal analysis or high-stakes financial decisions, you still need the human expert. Knowing that line is the difference between using AI intelligently and trusting it where it doesn't belong.
Which specific tasks in your work could you delegate more to AI today, and which ones should still have you making the final call?
Want to use these tools? See the unbiased reviews or back to the news.