A recent study published by Microsoft indicates that current AI systems face substantial difficulties with long-term decision-making. Researchers evaluated the performance of various models in tasks involving interconnected decisions over a simulated one-year period.
The findings show that AI models reached only 27% of human performance levels in these tests. The Qwen3.7-Max model, integrated with Hermes, demonstrated the best results among the tested systems but still failed to meet human benchmarks for long-horizon reliability.
The report highlights that AI models struggle specifically with scenarios requiring continuous self-correction and agency. These limitations suggest that current architectures have not yet mastered complex tasks that demand sustained strategic planning over extended timeframes.
Market participants are now watching how major developers, including Anthropic, Google, and Meta, will address these findings. Future model releases from these companies are expected to influence market sentiment and potentially alter the competitive landscape in the AI sector.
