xAI launches Grok 4 Fast — faster, cheaper unified reasoning model
xAI has announced Grok 4 Fast, a new variant of its Grok 4 family optimized for speed and cost-efficiency. According to xAI, Grok 4 Fast delivers similar performance to Grok 4 while using roughly 40% fewer “thinking” tokens on average and reducing cost to achieve the same frontier-benchmark performance by about 98%.
Key highlights
- Unified architecture: One model can operate in a deeper “reasoning” mode or a quicker “non-reasoning” mode via system prompts.
- Efficiency: ~40% fewer tokens used vs. Grok 4; xAI claims ~98% cost reduction to match Grok 4 on frontier benchmarks.
- Large context: Reported to support very large contexts (reports mention up to a 2M token context window).
- Benchmarks: Ranked #1 in LMArena’s Search Arena and #8 in the Text Arena in side-by-side comparisons.
- Availability: Rolled out to all users (including free tier) on web, iOS and Android.
Why it matters
Grok 4 Fast targets use cases that need high-throughput, low-latency responses—search, quick Q&A, browsing and coding tasks—by trading the heaviest reasoning passes for a smarter, token-efficient approach when possible. That cost-efficiency could make large-scale deployments much cheaper and more responsive.
Sources & further reading
- Official xAI announcement: https://x.ai/news/grok-4-fast
- Technical model card (PDF): Grok 4 Fast model card
- Community/API access: OpenRouter — Grok 4 Fast
Takeaway
Grok 4 Fast is positioned as a cost- and latency-optimized model that keeps competitive performance while lowering the token and dollar cost of inference. With competitors like Google (Gemini) and Anthropic (Claude) continuously updating, expect rapid follow-ups across the industry.
What are your thoughts on Grok 4 Fast and the race to build cheaper, faster LLMs? Share your view in the comments below.