The young shoot from former Google DeepMind and Anthropic details in-house models used for a coding assistant, and says it has automated a large part of the training cycle via agents. It announces a $1 billion Series A round, co-led by Kleiner Perkins and Clearlake, with participation from Google.
Homemade templates, a coding wizard and sizes up to 10 trillion
The Mirendil-Base model serves as the foundation for Fern, a coding model unveiled last month. According to the company, Fern is designed to automate programming challenges encountered in AI laboratories and can process 10 million tokens per prompt, which is, according to Mirendil, 10 times more than Claude Opus 5.5, which equates to approximately 500,000 lines of code. Mirendil says that Fern is based on a version of Mirendil-Base of between 100 and 500 billion parameters and that it presented, late last month, two improved versions peaking at up to 10,000 billion parameters. Earlier this year, the company commissioned a single researcher to create a coding model aimed at the level of leading open source LLMs. Mirendil reports that its automation made it possible to carry out this project solo, at a cost of $10 million and over approximately 3.5 months. The company says its agents wrote more than 99% of Mirendil-Base’s code, set up the infrastructure, chose the training algorithm, collected the data, and refined both the dataset and architecture.
AI strategies, movements and tricks deciphered, every evening in 5 minutes. Free.
Included upon registration: our selection of the best AI guides & comparisons.
Choose your pace
Free · No spam · Unsubscribe in 1 click
Automate training with Lor and hundreds of agents
The first product, Lor, automates many tasks related to training sessions. Mirendil explains that this automation relies on the distribution of work between hundreds of AI agents. The development of language models proceeds by testing variants and comparing the results; the company says Lor supports this flow, and the platform automatically corrects errors found during reviews.
Founders from DeepMind and Anthropic
Created last year, the company was co-founded by two former researchers from Google DeepMind and Anthropic PBC. Behnam Neyshabur, who serves as managing director, is among the inventors of the SAM algorithm, developed to optimize the performance of models on tasks for which they have not been explicitly trained. CTO Harsh Mehta worked on pre-training projects in both labs, a phase that is the initial stage of the AI ​​development workflow. Mirendil positions itself as a provider of AI R&D acceleration software.
A $1 billion round and stated ambitions
The company announces a $1 billion Series A round, released Thursday. Kleiner Perkins and Clearlake Capital co-led the investment, joined by more than half a dozen players, including Google LLC. Mirendil plans to use the presented algorithms to develop more efficient language models.

