A small San Francisco-based AI research startup is making waves in the tech world by releasing an unconventional artificial intelligence model that promises unprecedented user control and minimal content restrictions.

Nous Research, a relatively unknown player in the AI landscape, has introduced Hermes 4, a family of large language models that aim to outperform major tech companies’ AI systems. Unlike traditional AI models from OpenAI, Google, or Anthropic, Hermes 4 is designed to respond to almost any request without the typical safety guardrails.

The most striking feature of Hermes 4 is its “hybrid reasoning” capability, which allows users to toggle between quick responses and deeper, step-by-step thinking processes. In technical benchmarks, the largest 405-billion parameter model scored impressively high on mathematics and reasoning tests, rivaling systems that cost millions more to develop.

What sets Nous Research apart is its philosophy of prioritizing user agency over corporate content policies. The company believes that transparency and user control are more important than rigid safety constraints. This approach has made them popular among developers and researchers seeking maximum flexibility in AI interactions.

The model was trained using innovative systems like DataForge, a graph-based synthetic data generator, and Atropos, an open-source reinforcement learning framework. This unique training process required 192 Nvidia GPUs and 71,616 GPU hours, demonstrating that smaller startups can compete with tech giants through specialized techniques.

Nous Research’s release comes at a critical moment in AI development, challenging the notion that advanced AI capabilities should be controlled by a handful of corporations. By making the model weights freely available and offering various access methods, they’re pushing for a more open and democratized approach to artificial intelligence.

As the AI landscape continues to evolve, Hermes 4 represents a bold statement about the future of technology: innovation can emerge from unexpected places, and user empowerment might be the key to unlocking AI’s true potential.