Distilled reasoning model based on Llama 3.3 70B using DeepSeek R1 outputs.
Add a second model to compare.