Home/AI agents/Fern 0.1 (Mirendil)
Fern 0.1 (Mirendil)
Mirendil's first model, pretrained from scratch, with a context window of up to 10 million tokens. Access is early-access only.
Mirendil announced Fern 0.1 on October 26, 2026. It is pretrained from scratch. The context window is up to 10 million tokens. Mirendil says that is about 23 MB of code, about half a million lines. Fern 0.1 is Mirendil-Base (0.1-0.5T parameters) after a short round of post-training. Input is text and code. It is built for high-volume agentic tasks: agentic coding, debugging training runs from logs, reading agent traces end to end, and auditing training data. Access is listed as coming soon, with a request for early access. No pricing is published. Mirendil's own benchmark chart shows Fern at 6.1% on Terminal-Bench 4.0, so it is not a frontier coding model yet. Mirendil says more long-context results are coming soon.
Where it ranks
21 in AI agents
High-volume agent tasks, once early access exists.
See the ai agents list23 in Vibe coding
A long-context coding model, once early access exists.
See the vibe coding list
Pricing
Check site
No pricing is published. Access is listed as coming soon.
Checked
Added October 27, 2026
Last verified October 27, 2026
Dated releases
Mirendil announced Fern 0.1, its first model, with a context window of up to 10M tokens. Access is early-access only.