Flagship reasoning model with adaptive tool-use -- intelligently invokes retrieval and code interpreter on demand during inference. Advanced test-time scaling via RL.