Mirendil, an AI technology startup developing software to accelerate artificial intelligence development, has raised $1 billion in Series A funding in a round jointly led by Kleiner Perkins and Clearlake Capital. Google and several other investors also participated in the financing.
Founded last year by former Google DeepMind and Anthropic researchers, Mirendil aims to simplify and automate the process of building and training advanced AI models.
The company is led by CEO Behnam Neyshabur, a co-inventor of the Segment Anything Model (SAM), and CTO Harsh Mehta, who previously worked on AI pretraining projects at Google DeepMind and Anthropic.
Mirendil’s main product, called Lor, is an AI development platform that automates several tasks involved in training and improving AI models. It uses hundreds of AI agents to handle activities such as writing code, setting up infrastructure, collecting training data, testing algorithms, and fixing errors during development.
Building large language models (LLMs) typically requires extensive experimentation, with researchers testing different algorithms and model configurations to improve performance. Mirendil aims to reduce this manual work by allowing AI agents to manage much of the development process.
The company recently demonstrated this technology by developing its own coding model, Mirendil-Base. According to Mirendil, a single researcher completed the project in approximately 3.5 months at a cost of $10 million, with AI agents writing more than 99% of the model’s code.
Mirendil-Base also serves as the foundation for Fern, a specialized coding model introduced by the company last month. Fern is designed to help AI research laboratories automate complex programming tasks and can process up to 10 million input tokens per prompt, according to the company.
The company has also previewed upgraded versions of Mirendil-Base, with model sizes ranging from 500 billion to 10 trillion parameters, as it works toward developing more advanced AI systems.
With this $1 billion investment, Mirendil aims to advance its AI development platform and build more capable models by automating the complex processes involved in AI research and training.
