CollectivIQ Consensus Engine Outperforms Frontier AI Models on Benchmarks

The independent report from Ten Point Data highlights that the system reached a 53.3% accuracy rate on Humanity’s Last Exam, the highest text-only score identified in the study. Beyond raw accuracy, the platform demonstrated superior calibration, recording an Expected Calibration Error of 0.41. This marks a 28.1% improvement over the 0.57 error rate common among major standalone models, signaling a shift toward more reliable enterprise decision-making tools.

John Davie, CEO of CollectivIQ, noted that the architecture validates the principle that aggregating multiple AI perspectives yields higher-quality, lower-risk intelligence. Because the platform uses previous-generation models to outperform current flagship systems, it offers a cost-effective alternative to relying solely on the most computationally expensive technology. As businesses face an estimated $67.4 billion in annual losses due to AI hallucinations, this consensus-driven approach aims to reduce the four hours per week employees currently spend verifying inaccurate machine outputs.