Core Insights - xAI has launched a new AI model, Grok 4 Fast, which achieves reasoning performance close to Grok 4 while reducing average reasoning tokens by 40% and costs by 98% [1][8]. Performance Metrics - In the AIME 2025 no-tools test, Grok 4 Fast achieved a correctness rate of 92.0%, surpassing Grok 3 Mini and showing significant advantages in mathematical reasoning tasks like HMMT 2025 [3][5]. - Grok 4 Fast scored 85.7% in GPQA Diamond, 92.0% in AIME 2025, and 93.3% in HMMT 2025, demonstrating its competitive edge in various benchmarks [5]. Search and Information Retrieval - Grok 4 Fast exhibits advanced multihop search capabilities, ranking first in the LMArena search arena with an Elo score of 1163, leading the second place by 17 points [5][6]. - Its accuracy in tasks such as Chinese search and cross-platform data integration is significantly higher than that of comparable models [5]. Architectural Innovations - Grok 4 Fast integrates long-chain reasoning and fast response modes within the same model, allowing dynamic switching through system prompts, which reduces latency and further lowers token costs [7]. - Developers can adjust reasoning depth via the xAI API to meet various requirements, making it suitable for real-time search and code execution [7]. Accessibility - Grok 4 Fast is now available to all users, including free users, and is offered for free on platforms like OpenRouter and Vercel AI Gateway [8].
马斯克 xAI 最具性价比模型 Grok 4 Fast 登场:价格降低 98%
Sou Hu Cai Jing·2025-09-20 03:35