Hermes Agent v2026.7.20 significantly improved performance and reasoning capabilities, with new model options, smarter approval workflows, and enhanced agent coordination features.
Models and reasoning
- Mixture-of-Agents model: Select alongside Claude and GPT with visible reasoning from each reference model before the final answer streams.
- New model support: Added GPT-5.6, grok-4.5, Claude Sonnet 5, and reasoning effort tiers for flexible inference control.
- Live reasoning streaming: Agent reasoning now streams in real time instead of waiting for completion.
Workflow and automation
- Smart approvals: Independent LLM assessment replaces manual review of agent decisions.
- Completion contracts: Agents verify their own work by running your project's checks instead of guessing.
- 80% faster cold-start: Dramatically reduced initialization time for quicker agent launches.
- Live subagent transcripts: Watch delegated subagents work with durable background delegation and live visibility.
Interface and integrations
- New commands: /learn distills reusable skills, /journey views accumulated memories, /prompt composes in your editor, and /subscription manages Nous billing.
- Secret management: Plug Bitwarden and 1Password directly into Hermes for secure credential handling.
- Session export: Export conversations as Markdown, HTML, or Hugging Face traces for sharing and analysis.
Several other features were also released this month but are not listed here.