Mirendil is a pioneering AI research laboratory dedicated to democratising access to frontier AI research and development, thereby accelerating scientific and technological progress. The company's core mission revolves around building sophisticated AI systems that are exceptionally proficient at conducting AI R&D themselves. This involves not only training advanced, proprietary models from scratch, such as their Fern model, but also fundamentally redesigning the entire research laboratory workflow around these intelligent agents. The objective is to create a fully integrated, faster, more capable, and increasingly autonomous R&D loop.
At the heart of Mirendil's offering is Lor, an ambitious project designed to function as an entire AI lab. Lor's agents are engineered to write code, execute complex experiments, and iteratively refine results over extended periods, significantly reducing the human effort and time traditionally required for AI development. This agentic approach allows for rapid prototyping and exploration of new AI paradigms. Their first major model, Fern, exemplifies this strategy; it was pretrained from scratch and boasts an impressive context window of up to 10 million tokens, showcasing Mirendil's capability to build large-scale, high-performance foundation models with remarkable efficiency.
Mirendil's innovative methodology has enabled them to achieve significant milestones, such as pretraining a complex model like Fern with a single researcher and minimal compute resources (under $10 million) within a few months. This demonstrates a highly efficient and scalable approach to AI model development, which is a key differentiator in the competitive AI landscape. By empowering individual researchers and engineers with advanced AI tools, Mirendil aims to lower the barriers to entry for cutting-edge AI research, fostering a new era of innovation across various scientific and technological domains. Their work is particularly relevant for organisations and individuals looking to push the boundaries of AI without the need for vast, traditional research teams.