Jerusalem, 5 February, 2026 (TPS-IL) — Israeli AI startup Caura.ai unveiled PeerRank, a new autonomous framework in which AI models evaluate each other without human supervision. The system generates tasks, answers them using live web access, and produces bias-aware rankings.
PeerRank was tested across 12 commercial AI models, including GPT-5.2 and Claude Opus 4.5, generating over 253,000 peer-to-peer judgments. Research shows peer evaluation outperforms self-assessment, reliably identifying accuracy while exposing systematic biases in AI judgment. Claude Opus 4.5 narrowly topped the rankings in a blind evaluation.
Caura.ai co-developed the framework with Ben-Gurion University of the Negev, aiming to improve transparency and real-world assessment of AI performance.