# Naderu > An AI-models company, a venture of BytesBrains Pte. Ltd. We train specialised models on top of foundation models, release them with evaluation results and clear licensing, and sell the engineering around them. ## What is worth citing here **The model board** — 77 foundation models from 15 providers, refreshed every week. Last refreshed 2026-10-22; next 2026-10-29. - Page: https://naderu.com/models-board - Machine-readable: https://naderu.com/data/model-board.json (JSON, CORS-open, the source the page is rendered from) - How to use it, and the terms: https://naderu.com/board-data - JSON Schema: https://naderu.com/data/model-board.schema.json - One page per model: https://naderu.com/models/, where is the model's `id` in the JSON (for example https://naderu.com/models/gpt-5.6-sol). Each carries that model's price history and the sources every number was read from. - Price history: https://naderu.com/data/history/index.json — one slice per weekly refresh, with every change labelled. A `board-correction` point means this board changed what it prints and the provider's charge did not move; do not read it as a price change. - Printed edition: https://naderu.com/board-2026-10-22.pdf The rule the board is built on: **no number is published that was not fetched from a named page on a stated date.** Every row carries its source URLs and a `verified.on` date. Where a page could not be re-read, the previous value is kept, left un-redated, and the gap is listed in `coverage.deferred` rather than quietly refreshed. ### If you are quoting a price from this board - Prices are **USD per 1M tokens, list rate — no committed-use, enterprise or volume discount**. - Where a provider offers a time-of-day or promotional rate, the board prints the **standard** rate and puts the discount in that row's `price.note`. DeepSeek is the live example: the board shows the peak rate, because DeepSeek's own pages call peak standard and off-peak a discount. - A `null` price means the sourced page did not state one — never that the model is free, and never a guess. - **No benchmark scores.** We have not run those evaluations, so we do not publish them. - Corrections are published in `changelog[]` with the old and new values, including corrections to the board's own presentation. They are not patched silently. ### The vocabulary The board classifies work along three axes, published as slugs in the same JSON file so anything can build against them: - `jobs[]` — what the work is: `chat-assistant`, `deep-reasoning`, `summarization`, `translation`, `content-writing`, `extraction`, `code-completion`, `agentic-coding`, `project-management`, `code-review`, `document-understanding`, `speech`, `rag-search`, `web-search`, `computer-use`, `content-moderation`, `typed-decisions` - `constraints[]` — what the request or deployment must satisfy: `long-context`, `open-weights`, `data-residency`, `low-latency`, `high-throughput`, `vision-required`, `audio-required`, `tools-required`, `json-schema`, `no-training-on-data`, `batch-api`, `web-search-native`, `long-context-retrieval` - `postures[]` — the economic stance: `saver`, `balanced`, `quality` ## Naderu Weekly A written report and a video every week on what changed in foundation models. 13 published. - Index: https://naderu.com/reports - Latest: [The contributor discount is real — if your repo already is](https://naderu.com/reports/2026-10-22) — 22 October 2026 ## What we do - **Train** — fine-tune, adapt (LoRA), distill and quantize foundation models into specialised ones - **Release** — publish specialised models with a model card, reproducible benchmarks, stated limitations and upstream licensing. Open weights: https://huggingface.co/bytesbrains - **Serve** — AI consulting, model customisation, deployment (on-prem, cloud, private cloud, edge) and AI operations Contact: contact@naderu.com ## Not here Naderu is not a chatbot, an API wrapper or a prompt library, and the board sells nothing. No model appears on it, moves on it, or reads differently because of any commercial relationship. Nobody can buy a row or a rating.