#tool-use reinforcement learning

'xAI released Grok-4-Fast, a single prompt-steerable model with a 2M-token window and tool-use RL that reduces token usage by about 40% while maintaining frontier benchmark performance.'