CollectivIQ Tops Leading Frontier Models With 96.4% GPQA Diamond Accuracy Score According To Independent Benchmark Results; Validates Consensus AI Approach

BOSTON, Aug. 5, 2026 /PRNewswire/ -- CollectivIQ, the world's first AI consensus platform for business intelligence, today announced benchmark results showing its proprietary reasoning system beats massive frontier model giants for AI accuracy. CollectivIQ uses its proprietary algorithm to query the leading LLMs, including ChatGPT, Gemini, Claude, Grok and up to ten others at a time. The platform then highlights agreements and discrepancies to deliver the most accurate and truthful answer.

CollectivIQ is architected to optimize for cost per intelligence and to be maximally honest, systematically overcoming the sycophancy, overconfidence, and fabrication errors common in singular model architectures.

According to an independent report prepared by Ten Point Data this month, CollectivIQ achieved:

• 53.3% on Humanity's Last Exam (text-only): The highest published text-only score identified in the report, outperforming industry-standard frontier models.
• 96.4% on GPQA Diamond: #1 globally among published comparators and 26 percentage points above the human PhD baseline.

Beyond raw accuracy, the evaluation focused on calibration, or the ability to accurately signal uncertainty rather than confidently asserting inaccuracies. For Humanity's Last Exam, CollectivIQ recorded an Expected Calibration Error (ECE) of 0.41. This represents a 28.1% reduction in calibration error compared to standard major frontier models, which typically trend at a much higher ECE of 0.57. This ensures CollectivIQ remains a reliable partner for high-stakes enterprise decision-making. Notably, these benchmark results were achieved using previous-generation frontier models rather than today's latest flagship reasoning systems, demonstrating that CollectivIQ's consensus architecture can outperform cutting-edge single-model approaches without relying on the newest or most computationally expensive LLMs.

"I am incredibly excited to show that the old adage 'two minds,' or in our case, 'many minds, are better than one' proves out in AI, as it does in the real world," said John Davie, CEO of CollectivIQ. "By fusing responses from leading AI models through our proprietary consensus engine, CollectivIQ delivers higher-quality intelligence that is significantly more cost-effective and carries a fraction of the fabrication risk found in standard single models."

Existing data proves that AI hallucinations and fabrications are extremely costly and disruptive. For example, instances where models generate factually incorrect outputs are estimated to cost businesses $67.4 billion globally. Employees are now losing the equivalent of 51 workdays a year to technology friction, up 42% from 2025, with more than four hours a week spent just verifying AI outputs.

CollectivIQ is optimized as an asynchronous expert employee, leveraging the right LLM for the right level of cost and intelligence. Simple questions and tasks route to fast, inexpensive models, and hard, critical tasks route to the latest frontier models with the deepest reasoning. It prioritizes factual integrity over instantaneous chat speeds for every worker. It is designed for enterprise environments where the cost of a wrong answer is high and precision is paramount.

About CollectivIQ
CollectivIQ is the intelligence layer for the AI era, unifying leading LLMs into a single consensus-driven platform that delivers more accurate, secure, and cost-efficient insight. Founded by the leadership team behind Buyers Edge Platform, CollectivIQ was built to bring accountability, governance, and economic alignment to enterprise AI adoption. Learn more at www.collectiviq.ai.

SOURCE CollectivIQ