• •Nous Research claims Hermes MoA virtual models achieve 8% higher performance than Opus 4.8 and 11% higher than GPT 5.5
• •Reported performance gains are derived from the Mixture of Agents (MoA) architecture, a design that leverages multiple models to improve output quality
• •Independent verification of these metrics remains absent, and the specific test benchmarks used to calculate these percentage improvements have not been fully disclosed

Nous Research released data suggesting its new Hermes Mixture of Agents (MoA) models outperform both Opus 4.8 and GPT 5.5. Unlike standard monolithic model releases, the MoA approach routes tasks across a tiered stack of smaller models to maximize efficiency. However, the lack of third-party benchmark audits leaves the exact methodology behind these performance claims unverified. If these improvements hold under broader testing, it may force a shift in how developers weigh multi-model architectures against large-scale single models.