Athena 1.1 · Preview
Our own model, built in the open.
Athena 1.1 writes drafts, explains code and concepts, and answers from sources Atlas checks. It runs on hardware we control.
- ~8Bparameters
- Qwen3-8Bopen-weight base, Apache-2.0
- 1×Atlas token rate, the lowest
- Pointhardware, not a third-party model API
What Athena does well.
Drafts and rewrites
Emails, announcements and notes, written in a clear, finished voice.
Clear explanations
Code and concepts, step by step. Math is checked with a calculator as it is written.
Answers from sources
Atlas looks up sources first. Details that don’t match them are rewritten before you see the answer.
The facts, plainly.
- Developed by
- Point Software Solutions, for Atlas AI
- Base model
- Qwen3-8B, an open-weight model (Apache-2.0), about 8 billion parameters
- Where it runs
- On hardware we control, not a third-party model API
- Mode
- Chat only. Athena does not use tools, even in Agent mode; for tool work, choose Andromeda.
- Cost
- 1× Atlas tokens, the lowest rate
- Availability
- Preview, with limited availability
Where Athena shows up.
- In the model menu: pick “Athena 1.1 · Preview” to chat with it directly.
- As a column in Compare, beside Claude, GPT and Gemini.
- In Auto on the Lite tier, for fact and small-talk questions from looked-up sources. Every such answer is checked first; one that fails is redone by the larger engines.
How it is trained.
We are careful about where Athena’s training data comes from.
- Open questions, self-written answers. Training questions come from openly licensed datasets. Answers are written by the base model itself and kept only when they check out, plus examples hand-written by the Atlas team.
- Never trained on other assistants’ output. Athena is not trained on text written by Claude, GPT or Gemini. Athena 1.0 was retired for that reason: part of its training data had been written by another company’s model.
- Never trained on your data. No user conversations or connected-account data are used to train Athena, or any other model.
- Measured before release. Each version is scored against its untuned base on a fixed set of questions covering facts, instructions, math, honesty and writing, and ships only on the score of the file people actually run.
Athena 1.1 is a compact preview model. For demanding work, Andromeda routes to larger engines. AI can make mistakes; review important answers and code before relying on them.
Put Atlas to work.
Atlas AI is in invite-only early access on the web and Windows.