Mirendil unveils Fern, its first model, which it says can hold up to 10 million tokens of context
The one-year-old lab reports 45% accuracy on its own multi-hop benchmark at full context length. Access has not opened yet.
Mirendil announced Fern on October 26. It is the lab's first model, pretrained from scratch, and the company says it supports a context window of up to 10 million tokens. The first version, Fern 0.1, went through a short round of post-training and is described as Mirendil's most efficient model for focused, high-volume agentic tasks. It accepts text and code and has between 0.1 and 0.5 trillion parameters. Access is listed as coming soon, with early-access requests open. According to Startup Fortune, Mirendil says 10 million tokens is roughly 23 megabytes of code, or about 500,000 lines. Startup Fortune notes that Google's Gemini 4 Argon and Anthropic's Claude Opus 5 offer 1-million-token windows. On a company-run benchmark, Mirendil reported 45% accuracy on multi-hop questions at the full context length. That result has not been independently tested. Very long context could let developers load whole codebases or long agent logs at once instead of relying on retrieval systems. Mirendil came out of stealth in June with a $200 million seed round at a $1 billion valuation, co-led by Andreessen Horowitz and Kleiner Perkins, with Nvidia also investing.