When OpenAI launched GPT-5 in August, it marketed the model as its “smartest, fastest, most useful” release yet. While some users complained it felt weaker at creative writing, others noted it excelled at practical tasks like coding and data analysis. That raised a curious question: if GPT-5 is so good at processing complex data, how would it perform as a stock picker?
To find out, a journalist handed GPT-5 $500 in real money and asked it to invest in up to five stocks with the goal of maximizing returns over six months. The AI took eight minutes to calculate and returned with a bold “Diversified High-Growth Portfolio.” Its choices: Palantir (PLTR), AppLovin (APP), Agios Pharmaceuticals (AGIO), Hut 8 Corp. (HUT), and MicroStrategy (MSTR), as detailed in Fast Company.
Each pick came with an explanation. Palantir was selected for its AI-driven government and commercial traction. AppLovin, the AI noted, had a proprietary engine and was trading just below peak levels, leaving room for growth. Agios was a risky bet on FDA approval for its drug Pyrukynd, which could open the thalassemia treatment market. Hut 8 and MicroStrategy were included as crypto plays, with MicroStrategy holding billions in Bitcoin and Hut 8 pivoting into energy infrastructure for AI and mining.
GPT-5 made it clear: this wasn’t a safe portfolio. It deliberately targeted volatile, high-upside sectors – AI, biotech, and crypto. In the model’s own words, it aimed for “explosive upside rather than stability.” Reuters has already reported the suprising performance of a ‘ChatGPT fund’ portfolio in a separate experiment.
Two weeks after the purchases, the portfolio was already up about 10 percent. That’s a promising start, but it highlights just how speculative the strategy is. Palantir’s continued momentum, biotech trial outcomes, and Bitcoin’s notorious volatility could easily swing results in either direction.
Financial experts caution against leaning too heavily on AI for investment decisions. Research suggests large language models can analyze patterns and summarize reports but often struggle to consistently outperform traditional models or even simple index funds. The GPT-5 picks demonstrate sharp instruction-following and clear reasoning, but whether that translates to lasting gains remains to be seen.
In the end, the experiment shows GPT-5 won’t shy away from rolling the dice when asked. The portfolio is risky, aggressive, and true to the user’s request. The real test will be whether those bets look like genius or folly six months down the line.
