Building intelligence for everything you do.
SORAEN is an independent AI research lab. We build models for language, reasoning, code, vision and voice, and we publish how they perform.
- Open weights
- Downloadable at release
- Public evaluations
- Losses included
- Licensed data
- Every source credited
- Safety tested
- Before every release
AI should earn trust the way science does: in the open, with evidence anyone can check.
One lab. Every kind of intelligence.
SORAEN Lite is where we start, not where we stop.
Language
Reading, writing and conversation in many languages
Larger SORAEN models
Reasoning
Step-by-step solutions to mathematical problems
A dedicated reasoning mode
Code
Python with explanations, plus tool calling
A specialised coding model
Vision
Photos, screenshots, charts and scanned pages
A vision-first model
Voice
Spoken questions through speech-to-text
Natural voice conversation
Safety
Safety training and red-team evaluation
Published tests for every release
SORAEN Lite
A capable all-rounder, small enough to run on a single GPU. It handles text, images and voice, with a particular strength in answering from your own documents, with the source of every fact.
- Parameters
- 9 billion
- Built on
- Qwen3.5-9B (Apache 2.0)
- Training data
- About 12,500 curated examples
- Release
- Only if it passes its evaluation
How long do refunds take?
Refunds are issued within 14 days of purchase [1]. To request one, contact the billing team from your account page [2].
Made for real work.
Answers from your documents
Upload files to a private collection. Keyword search and semantic search work together to find the right passages.
Citations on every fact
Each claim points to its source as [1], [2], and the response lists which document and passage each number refers to.
Says when it doesn't know
When none of your documents contain the answer, SORAEN Lite says so instead of inventing one. We train and measure this directly.
Simple REST API
One HTTPS endpoint, JSON in and out, from any language. Streaming, system prompts, tool calling and multi-turn chat are built in.
Images and voice
Ask about photos, screenshots, charts and scanned pages, or just speak your question instead of typing it.
Private by design
Each document collection belongs to one API key. No other customer can search it, list it or read it.
Build with SORAEN.
One API for conversation, documents, images and voice. Attach a collection and every answer arrives with its sources.
- Chat completions with streaming
- Document collections: upload, replace, search, delete
- Sources returned with every grounded answer
- Per-key rate limits and monthly usage you can check anytime
Trust is earned in public.
SORAEN is an independent AI research lab. We would rather show you our results than ask you to believe us, including the tests where a model gets worse.
Questions? You'll reach the founder directly: founder.soraen.ai@gmail.com
See how it's trained
The data mix, the filters and the full training recipe.
Check the results
Base model and SORAEN side by side, on the same tests.
Read our safety approach
What we test before release and check in production.
Know where your data goes
We don't train on your documents or conversations.
An AI that works the way you do.
Set how SORAEN addresses you, how it works and what it focuses on. Founding places are open, with nothing to pay today.
Request early access.
We're opening SORAEN in stages. Leave your email and we'll write to you when your access is ready.