Elon Musk’s xAI released Grok 4 Fast on Friday, a streamlined artificial intelligence (AI) model that costs 98% less to run than its predecessor while maintaining similar performance.
The new model processes queries using 40% fewer computational units than Grok 4. The model starts at $0.20 per million tokens, far below the $10-plus rates common in the AI industry.
“Grok 4 Fast sets a new frontier in cost-efficient intelligence,” xAI stated in its announcement blog post.
Source: xAI
Artificial Analysis ranked Grok 4 Fast as having the best price-to-performance ratio among publicly available models, operating 47 times cheaper than the original Grok 4.
Performance benchmarks show Grok 4 Fast competing effectively despite its lower resource requirements. The model topped LMArena’s Search Arena leaderboard with an Elo rating of 1,163, beating OpenAI’s o3-search by 17 points. In general text tasks, it ranks eighth globally.
xAI achieved these efficiencies through reinforcement learning techniques that optimize the model’s engagement in complex reasoning versus quick responses. The system automatically switches between modes depending on query difficulty, reducing unnecessary computation.
Two versions launched through the xAI API: grok-4-fast-reasoning for complex problems and grok-4-fast-non-reasoning for simple queries. Both support a 2-million-token context window, allowing analysis of extensive documents.
Free access continues temporarily through OpenRouter and Vercel AI Gateway. Regular pricing ranges from $0.20 to $1.00 per million tokens based on usage volume and type.
Competition intensifies as AI companies race to balance capability with affordability. Google’s Gemini 2.5 and Anthropic’s Claude Opus 4.1 represent similar efforts to optimize performance while controlling costs.
All users, including free accounts, can access Grok 4 Fast through grok.com, iOS, and Android applications starting Saturday. The democratized access marks a shift from typical AI pricing models that reserve advanced features for paying customers.
The company plans continuous updates based on user feedback. Future versions will add “enhanced multimodal capabilities and agentic features”.
The release followed major staff cuts on September 12. xAI terminated its entire data annotation team, the department that trains AI models. These 500 workers learned about the layoffs through evening emails informing them that their salaries would continue through November 30, although system access would end immediately.