Epoch AI's InnovationEval benchmark shows current models lag human algorithmic problem-solving

According to Epoch AI, the InnovationEval benchmark demonstrates that recent AI models struggle to replicate human algorithmic breakthroughs, suggesting a measurable capability gap in novel problem-solving compared to human researchers. The benchmark score of 89 on Hacker News indicates current models have not yet matched human-level innovation in algorithmic domains. The finding suggests limits to agent capability for independent research and discovery tasks.