Advanced reasoning model that's 200% faster than R1-0528 with improved token efficiency and balanced speed-coherence performance.