Fern 0.1

View Blog

AI research model with up to 10 million tokens of context for coding, debugging training runs, and analyzing agent traces; early access through Lor.

Links

Built by

Mirendil

Cosigned by

No cosigns yet

Fern 0.1 is Mirendil’s first language model, based on Mirendil-Base after a short round of post-training. Mirendil says its architecture, training, and serving support a context window of up to 10 million tokens, with text and code input. The company presents it for high-volume agentic work in AI research and development, including auditing training data, debugging training runs from logs, working across large code repositories, and analyzing agent traces. The announcement describes benchmark results and long-context tests, while noting that more comprehensive long-context results are forthcoming.

Built with

PyTorch

Mirendil says its base-model pretraining project was implemented from scratch on top of standard libraries including PyTorch and JAX; Fern 0.1 is described as Mirendil-Base after post-training. This supports these as reported implementation dependencies of its model lineage, not as independently verified Fern-specific manifests.

JAX

Be the first Cosign

Put your name behind Fern 0.1.

Add it to your Products I use list. You can also share why you use it.

Product name

Story

Product name

Story

Product name

Story

Product name

Story

Product name

Story

Fern 0.1

Built with

PyTorch

JAX

Put your name behind Fern 0.1.