@InclusionAI This is the way to go.
As of today this is the best and most efficient model I have ever seen.
If you want my advice: you should continue working in this direction.
We don't need models that "know everything" but models able to reason correctly.
A model able to reason (same or better than us) will be able to help on many different fields, given the right data in the context.
Your model is the first ever small model to be able to solve one particular problem (that I have never published and is not in any dataset).
In the past (2024), even models like gemini flash or pro could not solve it.
Today most big models can solve it and even gemini flash can.
But small models (including qwen 3.5 9B) still can't solve it.
Your model could solve it TWICE in a row.
This means that better reasoning can be achieved even with smaller models and this is the way to go
If you need any help, I am available to brainstorm the situation or help anyway I can.
Best regards.