Mirendil launches Fern 0.1, its first model with 10M-token context

Mirendil has released Fern 0.1, a from-scratch coding model with a context window of up to 10M tokens, roughly 23 MB of code. It has 0.1–0.5T parameters and is available only as hosted early access. The base model was pretrained by one researcher for under $10M, with AI agents writing over 99% of the code.

• Context up to 10M tokens, about 23 MB or 500K lines of code
• RULER single-needle retrieval: 100% from 32K to 10M tokens
• Terminal Bench 4.0: 6.1% versus 66.4% for Claude Opus 5.5
• Mirendil-Base pretrained in 3.5 months for under $10M