A compact, no-frills text model that punches at a reasonable level for its size. StabilityAI's 3B parameter model trained on 4 epochs of 1 trillion tokens, making it more data-efficient than many peers. It handles straightforward language tasks adequately but shows its limits quickly on complex reasoning or lengthy context — the 4096-token window and small parameter count mean it's better suited for focused, contained tasks than sprawling conversations.