This is an XL (~76M param) model trained on UltraChat, FineWeb-Longish, and Codex. It supports Thinking, Tool Calling, and Advanced Reasoning.
