The 800-million-parameter model outperformed Qwen 9B and Gemma 4 in generating Linux commands

The specialized dale-0.8B achieved 79.4% on a test set of 471 tasks—significantly outperforming general-purpose models that are 5 to 11…