Lab · Jah 7.0

Jah 7.0. Our own AI model. Public weights. Private work.

The best public AI handles public work. Anything private goes to Kiyomi, our private AI for offices with sensitive files. Kiyomi runs on Jah 7.0 and other zero-retention models, and anyone can download Jah's weights and check them.

For your business

Have the RMDW team put it to work

Private AI for the work that can’t go into a public chatbot: client files, contracts, records. Set up and kept running by the RMDW team.

START WITH THE ASSESSMENT >For builders

Plug it into your own tools

One-click setup prompts for Claude Code, Codex and Cursor, plus the endpoint for any OpenAI-compatible tool.

SET UP JAH >

RMDW Bench v1 · 20 real coding tasks

Jah 7.0 (RMDW)19/20

Claude Opus 519/20

GPT-5.619/20

Claude Fable 518/20

Gemini 3.718/20

Kimi K317/20

19/20 is the ceiling: one task is impossible on purpose

How we test, and the hard suite

What Jah 7.0 can do

Reads: Very long documents in one go (about a million tokens)
Sees: Text and images
Thinks: Answers fast, or reasons first when the job is hard
Works with tools: Calls tools, and works inside Claude Code, Codex and Cursor
Open: The weights are public on Hugging Face

Why we built our own

Business owners don’t ask which model we use. They ask two things: can it get the job done, and can it stay private? Doctors, lawyers and financial firms ask the second one first.

So the RMDW team splits the work. Public work gets the best public AI there is. Private work goes to Kiyomi, which runs on Jah 7.0, a model anyone can inspect, and other zero-retention models.

Read: Why we run Claude Opus 5.5 and our own Jah 7.0 The weights on Hugging Face ↗