xAI Unveils Grok4Fast: 40% More Efficient Than Grok4
xAI Launches High-Efficiency Grok4Fast Model
xAI has introduced Grok4Fast, a new lightweight flagship AI model that delivers comparable performance to its predecessor Grok4 while requiring 40% less computational power. According to company reports, this efficiency improvement could reduce task costs by up to 98%.
Performance Benchmarks Show Competitive Edge
The model demonstrates impressive results across multiple benchmark tests:
- 85.7% on GPQA Diamond
- 92.0% on AIME2025
These scores place Grok4Fast in competition with top-tier models like Grok4 and GPT-5. xAI attributes this performance to optimized "thinking tokens," achieving similar results with significantly fewer tokens than previous versions.
Innovative Architecture Design
Breaking from traditional approaches, Grok4Fast features:
- Integrated architecture combining multiple approaches
- Behavior control through system prompts
- Strong external tool capabilities including web browsing and code execution
The model outperforms Grok4 in benchmark tests like BrowseComp and X Bench Deepsearch, even surpassing OpenAI's o3-websearch model in the LMArena-Search benchmark.
Availability and Pricing Structure
Grok4Fast offers two specialized versions:
- Inference-intensive task optimization
- Quick answer focus version
Both support a context window of 2 million tokens and are available through:
- grok.com website iOS/Android apps xAI API Pricing ranges from $0.05 to $1.00 per million tokens, with free access currently available via OpenRouter and Vercel. ## Key Points:
- 40% efficiency improvement over Grok4
- Matches top model performance at reduced cost
- Integrated architecture replaces separate task models
- Superior tool usage capabilities
- Competitive pricing starting at $0.05/million tokens
