xAI today released Grok 4 Fast, a reasoning model that trades a small fraction of raw accuracy for dramatic cost savings. The company says the model uses about 40 percent fewer thinking tokens on average than its flagship Grok 4, translating to an estimated 98 percent lower cost per benchmark point.

Grok 4 Fast unifies reasoning and non-reasoning behavior in a single set of weights steered by the system prompt, rather than requiring two separate model variants. It carries a 2-million-token context window and was trained end-to-end with reinforcement learning for tool use, including web browsing and X search. Reported benchmark scores include roughly 92 percent on AIME 2025, 85.7 percent on GPQA Diamond, and 80 percent on LiveCodeBench — close to Grok 4’s numbers.

The model ships in Grok’s Fast and Auto modes on web and mobile, and for the first time reaches free users. On the API, it is exposed as grok-4-fast-reasoning and grok-4-fast-non-reasoning. Early community reaction fixates on its cost-to-intelligence ratio, with independent trackers rating it highly for intelligence per dollar.

xAI frames the release as a bet that the frontier race is shifting toward efficiency, not just capability. Whether developers and users agree will determine if ‘fast and cheap’ becomes as coveted as ‘smartest in the room.’