April 2026 Summaries

The Trinity Builders Program is a community credit grant initiative designed to provide developers, researchers, and open source builders with free inference access to Trinity models via the company's API. This program aims to support projects that involve Trinity models, especially those that push the boundaries of reasoning models, extend the Trinity ecosystem, or require significant compute resources to progress. By offering free API credits, the program seeks to empower participants to experiment, iterate, and potentially develop products with Trinity models, while also gathering valuable feedback to guide future model developments. The credits, allocated based on project scope and available capacity, are valid for 90 days, and the program is open to a wide range of projects, with applications reviewed on a rolling basis. For those unable to secure a grant, Trinity models are also available in quantized variants suitable for running on consumer hardware, ensuring broader accessibility.

Running a high-reasoning AI assistant on personal hardware is now more accessible with the Hermes agent, which can be set up on operating systems like Linux, macOS, and Windows. The setup process involves installing the Hermes agent with a simple command, which creates a virtual environment and installs necessary dependencies. Users go through a setup wizard to select their model and provider, configure the terminal backend, adjust agent settings, and optionally link messaging platforms such as Telegram or Discord. The agent supports various tools like web search, browser control, and code execution, enhancing its functionality. The walkthrough provides guidance on configuring the Trinity Large model, managing context settings, and setting up browser automation and text-to-speech features. After completing the installation, the agent is launched, showing all tools and skills ready for use, with the status bar displaying the correct model configuration, confirming the successful setup of the AI assistant.

Trinity-Large-Thinking, a newly released reasoning model under the Apache 2.0 license, represents a significant milestone for the company, following months of development aimed at building serious American open models that developers and enterprises can own. This model evolved from smaller versions and a Preview model launched in January, which proved popular and effective in real-world applications, helping refine its capabilities. Trinity-Large-Thinking enhances multi-turn tool usage, context coherence, and instruction following, offering a robust and cost-effective solution for developers and enterprises. With its release, the model is available through an API and on Hugging Face, aiming to provide reliable, high-quality agents while maintaining a low price per token. The development journey involved overcoming technical challenges and optimizing training processes, with lessons from this model expected to improve smaller models like Trinity-Mini and Nano. The company's focus remains on delivering efficient, open-weight models that support the U.S.'s leadership in AI, with ongoing plans to enhance this generation of Trinity-Large.