What we're building next
Our roadmap, in the order we expect to ship it. Dates are announced when a release is ready, not before.
SORAEN Lite training
Two-stage training on the curated data mix, followed by the full benchmark run against the base model.
Public evaluation report
Every benchmark score for SORAEN Lite and its base model, published together, losses included.
SORAEN API launch
API keys for everyone on the waitlist, monthly usage tracking and per-key rate limits.
SORAEN Personal
Your own profile, tone and focus, applied to every conversation. Reservations are open.
Chat in the browser
A free chat app for SORAEN Lite, with sign-in, saved conversations, images and voice.
Document collections
Upload files, then ask questions answered only from them, each fact linked to its source.
Open weights
SORAEN Lite available to download and run on your own hardware.
Specialised models
Separate models for reasoning, code, vision and speech, each with its own public evaluation.
Larger SORAEN models
Bigger models trained the same way and judged by the same public tests.
Reasoning mode
Training for step-by-step thinking on hard problems, switchable per request.
Self-serve plans
Sign up, choose a plan and pay online, with usage shown in a dashboard.
Python and JavaScript SDKs
Official client libraries for the SORAEN API.
What's shipped
SORAEN website launched, with the waitlist open.
SORAEN Lite moved to the newer Qwen3.5-9B base, with a focused data mix of about 12,500 examples, safety training and a red-team benchmark.
SORAEN Lite training started.
SORAEN API built: keys, rate limits, usage metering and document collections with cited answers.
Hear about each release first
Waitlist members get access as soon as each item ships.