Naderu — specialised models by BytesBrains

We train foundation models into specialised ones, release them with evidence, and sell the engineering around them.

Naderu is a venture of BytesBrains Pte. Ltd. — naderu.com.

How we release

No model without evaluation. A model is published here only after it passes its evaluation gate, and its card cites the exact eval suite version.
Provenance stated plainly. Every model names the upstream foundation model it was built from, and the terms that come with it.
Reproducible. Reported numbers come from a versioned training recipe and a versioned eval suite.

Model	What it is for	Built from
`naderu-laya-150m`	Fast CPU/edge decision routing — support, billing, sales, offboarding	`answerdotai/ModernBERT-base`
`hunter-crypto-7b`	Cryptography attacks for authorized security and CTF work	`Qwen/Qwen2.5-Coder-7B-Instruct`
`naderu-mini-4b`	On-device offline chat, GGUF	`Qwen/Qwen3-4B`
`naderu-loom-7b`	Offline interactive-fiction narrator with structured turns	`mistralai/Mistral-7B-Instruct-v0.3`
`naderu-geek-py-0.5b`	Small Python coding model	`Qwen/Qwen2.5-Coder-0.5B-Instruct`

Evaluation results, limitations, and licensing are in each model card.

Model board — a weekly, sourced comparison of the models that matter, with a printed edition.
Weekly reports — what changed and why.
Services — fine-tuning, customization, deployment, and operations for teams who want a model they own. naderu.com