Mirendil releases Fern 0.1, its first model, with a 10M-token context window

Mirendil released Fern 0.1, its first model, which it pretrained from scratch with a 10M-token context window. The company positions the model for agentic AI R&D work.

On the RULER single needle test at 10M tokens, Fern 0.1 scores 100%. The same report says the model falls behind frontier models on agentic coding.

Developers building agents that need to work across very long inputs now have another 10M-token option to evaluate, though the reported results are limited to one retrieval benchmark.

Because the model trails frontier models on agentic coding, teams with coding-heavy agents should test it on their own tasks before switching any workloads to it.