A new benchmark named Taste-Bench has been introduced to evaluate AI agents' ability to make strategic decisions over extended tasks, according to The Decoder. This test focuses on how well AI models can navigate complex, long-running scenarios requiring thoughtful choices.
Results from the benchmark show that GPT-5.6 Sol correctly selects the optimal path in just under 60 percent of simple binary decision questions. Interestingly, The Decoder reports that allowing AI agents more time to think does not improve their accuracy, suggesting inherent challenges in current AI decision-making processes.
For Japanese investors and traders in FX, crypto, and equities, these findings highlight the current limits of AI-driven strategies, emphasizing the need for cautious integration of AI tools in market decision systems.