Hermes
#10 in Open-Source-Spraakmodellenous-research · siet 25. Februar 2026 · 51× · tolest 30. Juni 2026
53
Momentum
Hermes Agent is a self-improving, open-source AI agent by Nous Research, launched on GitHub on February 25, 2026. It features a built-in learning loop: after each task it creates reusable skill files, persists knowledge across sessions, and builds an evolving model of the user over time. The framework runs self-hosted (CLI, desktop, server), is model-agnostic (20+ LLM providers), supports 20+ messaging platforms, and is MIT-licensed with no token costs for the agent itself.
Momentum-Verloop
04.04.03.07.
Features
| Key Benchmark (%) | Internal metric: Agents with 20+ self-created skills solve comparable follow-up tasks 40% faster (lower token usage + wall-clock time) than fresh instances (Nous Research, April 2026) |
| Context Window (Tokens) | Model-agnostic; Hermes works with any LLM supporting 64K+ context; xAI Grok 4.3 via integration: 1M token context window |
| License | MIT License |
| Multimodality | Yes: vision/image analysis, image generation (FAL, Krea), text-to-speech (OpenAI TTS), speech-to-text (Whisper, Groq et al.), browser automation, video generation (pluggable) |
| Platform | macOS 12+, Windows 10/11, Linux (any distro), WSL2, Android/Termux (experimental); CLI, desktop app, web dashboard; 6 terminal backends (local, Docker, SSH, Daytona, Singularity, Modal) |
| Price | Agent free (self-hosted); optional Nous Portal: Free $0, Plus $20/mo ($22 credits), Super $100/mo ($110 credits), Ultra $200/mo ($220 credits) |
| Price per 1M Tokens | Model-dependent via provider; Nous Portal models from $0.02/1M input (DeepHermes 3 Mistral 24B) up to $1.00/1M input (upper limit for NousResearch models) |
| Release Date | February 25, 2026 |