Hermes MoA virtual models:8% higher than Opus 4.8, 11% higher than GPT 5.5

- Nous Research is an independent AI lab focused on open-source language model development, known for Hermes-series models built on top of base models like Llama or Mistral. - Mixture-of-Agents (MoA) is a technique where multiple language models collaborate — outputs from several "worker" models are fed into a "synthesizer" model to produce a better final answer, often improving reasoning and accuracy. - The tweet compares MoA-style Hermes models against GPT-4 (Opus-class) and a hypothetical "GPT-5.5" — these abbreviations refer to OpenAI's frontier models; higher percentages indicate better benchmark performance. - The claims (8-11% improvements over top closed-source models) are notable because open-source approaches rarely beat proprietary leaders on standard benchmarks, though such comparisons often depend on specific tests and setups.