New SOTA: TrustedRouter Fusion Beats Fable and Frontier (trustedrouter.com)

TrustedRouter has announced a breakthrough in AI model benchmarking by achieving a new state-of-the-art (SOTA) score of 70.6 on the DRACO benchmark, surpassing the previous best of 69.0 set by OpenRouter’s Fusion method. This significant achievement highlights the potential of open science in AI research, as TrustedRouter employs a fully open-source approach, providing transparency through verifiable results. By utilizing a diverse panel of models—including frontier open-weight models like DeepSeek V4 Pro and Kimi K2.6 alongside GPT-5.5 and Claude Opus 4.8—the new fusion technique harnesses differing model strengths to synthesize superior outcomes. Key to this success is the operational framework of TrustedRouter, which runs benchmarks within a Trusted Execution Environment (TEE) that ensures data privacy and integrity. This setup means that all model interactions, searches, and results are conducted under tight security, with no data leakage. The transparency of the entire process—from code to results—is designed to build trust within the AI/ML community, allowing other researchers to replicate the results independently. By prioritizing reproducibility and openness, TrustedRouter sets a new standard for AI benchmarking that could influence future research practices and model evaluations.