A short internal note for anyone watching the chess project.
What changed
The language layer in Rose Sensei was swapped today. The new model is faster on long games and gives more grounded commentary on positions where the engine evaluation is close.
What didn't change
- The strength layer is still Stockfish NNUE.
- The grounding layer and prompt rules are unchanged.
- The product behaviour, the UI, and the public API are all unchanged.
If you open a game review and the commentary reads tighter than last week, that's why. No user action required.