Pushing the Frontier of Cost-Efficient Intelligence

Grok 4 Fast
Pushing the Frontier of Cost-Efficient Intelligence

We're thrilled to present Grok 4 Fast, our latest advancement in cost-efficient reasoning models. Built on xAI’s learnings from Grok 4, Grok 4 Fast delivers frontier-level performance across Enterprise and Consumer domains—with exceptional token efficiency. This model pushes the boundaries for smaller and faster AI, making high-quality reasoning accessible to more users and developers. Grok 4 Fast features state-of-the-art (SOTA) cost-efficiency, cutting-edge web and X search capabilities, a 2M token context window, and a unified architecture that blends reasoning and non-reasoning modes in one model.

Advancing Cost-Efficient Intelligence
Grok 4 Fast sets a new frontier in cost-efficient intelligence, outperforming Grok 3 Mini across reasoning benchmarks while slashing token costs.

We used large-scale reinforcement learning to maximize the intelligence density of Grok 4 Fast. In our evaluations, Grok 4 Fast achieves comparable performance to Grok 4 on benchmarks while using 40% fewer thinking tokens on average.

This 40% increase in Grok 4 Fast's token efficiency, combined with a significantly lower price per token, results in a 98% reduction in price to achieve the same performance on frontier benchmarks as Grok 4. As verified by an independent review from Artificial Analysis, Grok 4 Fast exhibits a state-of-the-art (SOTA) price-to-intelligence ratio compared to other publicly available models on the Artificial Analysis Intelligence Index.

Native Tool Use with SOTA Search
Grok 4 Fast was trained end-to-end with tool-use reinforcement learning (RL). It excels at deciding when to invoke tools like code execution or web browsing.

For instance, Grok 4 Fast exhibits frontier agentic search capabilities, seamlessly browsing the web and X to augment queries with real-time data. It hops through links, ingests media (including images and videos on X), and synthesizes findings at light speed.

Frontier of General Post-training
Grok 4 Fast also establishes a new cost-effective frontier on general domain. We are excited to share Grok 4 Fast’s result on LMArena, where it has been privately battle-testing on the Search and Text Arenas.

In LMArena's Search Arena, grok-4-fast-search (code name: menlo) claims #1 with 1163 Elo — a commanding margin of 17 over o3-search. Its superior reasoning efficiency and intelligence density enable it to surpass much larger models on real-world, search-related tasks.

In LMArena's Text Arena, grok-4-fast (code name: tahoe) ranks #8, performing on par with grok-4-0709 and highlighting its remarkable intelligence density. Notably, it significantly outperforms peers in its weight class, where all comparable size models rank 18th or below.