**[Hermes Agent Launches MoA Functionality: Multi-Model Collaboration Outperforms Single Models]** According to monitoring by Beating, the open-source agent platform Hermes Agent has officially introduced support for the Mixture of Agents (MoA) preset. MoA has now transitioned into a virtual model provider and is no longer listed as a traditional underlying tool. The agent platform presents the MoA preset as a virtual model, allowing users to directly select it via the `/model` command or the model dropdown menu in the desktop application. In addition to switching via `/model`, users can also use the `/moa [prompt]` syntax for one-time quick calls. During a single conversation round, the system temporarily activates the default MoA preset, automatically reverting to the previous model after completing the response. In the MoA operational mechanism, the user-selected preset includes both reference models and aggregation models. The system first invokes the reference model, providing it with simplified dialogue text that excludes system prompts and tool invocation history. The reference model generates an analytical opinion, which is then appended to the end of the user's latest input. The aggregation model, acting as the main executor, receives the analytical opinion and generates the final response under the complete tool schema and system prompts, executing any necessary tool calls. In the upcoming HermesBench benchmark tests, the MoA preset using Claude-Opus-4.8 as the aggregation model and GPT-5.5 as the reference model achieved a score of 82.02%, representing a 6-point improvement (approximately 8%) over running Claude-Opus-4.8 alone and an 11% improvement over running GPT-5.5 alone. To ensure the efficiency of the prompt cache, the dialogue input for the reference model is simplified to maintain stable caching. The aggregation model appends the reference opinion to the end of the prompt, ensuring the byte stability of the historical dialogue prefix and avoiding disruption of the cached context. Users can view presets by running `hermes moa list` or add and modify presets using `hermes moa configure [name]`. In preset configurations, recursive nesting is strictly prohibited, and the aggregation model of an MoA preset cannot point to another MoA preset. If the reference model encounters errors such as credential invalidation, Hermes Agent will not interrupt the task but will pass the error information into the aggregation model's context and continue execution. The MoA mechanism is now simultaneously available on CLI, gateway, desktop, and TUI interfaces. [Original Link]