The top AI companies are continuing to push the frontier on the performance and cost of AI models.
xAI appears to have made a significant breakthrough in the AI model landscape with its Grok 4 Fast model. According to analysis from Artificial Intelligence, a company that benchmarks AI models, Grok 4 Fast achieves Gemini 2.5 Pro-level intelligence while costing approximately 25 times less to operate. Grok 4 also has a 2M context window, twice that of Gemini 2.5 Pro.
Intelligence Performance Matches Top-Tier Models
In pre-release testing conducted by Artificial Intelligence, Grok 4 Fast demonstrated impressive capabilities in reasoning mode, scoring 60 on the Artificial Analysis Intelligence Index. This performance places it on par with Google’s Gemini 2.5 Pro and Anthropic’s Claude 4.1 Opus, establishing it as a serious contender in the frontier AI model category.
The model particularly excelled in coding evaluations, claiming the top position on the LiveCodeBench leaderboard. Remarkably, Grok 4 Fast even outperformed its larger sibling, the full Grok 4 model, in these coding benchmarks, suggesting that xAI has achieved significant optimization in this specialized area.
While Grok 4 Fast performs below the flagship Grok 4 and OpenAI’s GPT-5 (high) models, its performance-to-cost ratio represents a major advancement in making high-quality AI reasoning accessible at scale.
Grok 4 Fast Speed and Pricing
The standout feature of Grok 4 Fast lies in its economic efficiency. xAI has priced the model at just $0.2 per million input tokens and $0.5 per million output tokens, making it one of the most cost-effective reasoning models available.
Beyond competitive pricing, Grok 4 Fast demonstrates superior token efficiency compared to its competitors. The model required only 61 million tokens to complete Artificial Intelligence’s intelligence index, significantly fewer than Gemini 2.5 Pro’s 93 million tokens and the full Grok 4’s 120 million tokens.
This combination of low pricing and high efficiency translates to dramatic cost savings for businesses. Running the Artificial Analysis Intelligence Index on Grok 4 Fast costs approximately 25 times less than using Gemini 2.5 Pro and 23 times less than GPT-5 in reasoning mode.
Performance benchmarking revealed that Grok 4 Fast is actually fast — during pre-release testing, xAI’s API endpoint achieved 344 output tokens per second, approximately 2.5 times faster than OpenAI’s GPT-5 API. The model also delivered impressive end-to-end latency results of just 3.8 seconds, making it faster than most non-reasoning models for many workloads. This speed advantage positions Grok 4 Fast as an attractive option for real-time applications where both reasoning capability and response time are critical.
However, Artificial Intelligence notes that these speeds may decrease as API traffic increases, and they recommend monitoring their live performance benchmarking for ongoing updates.
Market Implications
xAI is still looking for a breakout use-case for Grok that can catapult it into the mainstream. ChatGPT had its Ghibli moment, Google Gemini managed a similar feat with Nano Banana, but Grok’s bet on Imagine — its video creation tool — hasn’t quite worked out the same way. It now seems to be focusing on a more practical use-case — coding. Grok’s models have been topping leaderboards on OpenRouter for coding use-cases, and Grok 4 Fast’s pricing seems designed to take some of the coding market from other players, particularly Anthropic. It remains to be seen how Grok 4 is adopted by the community, but the model benchmarks do show potential of it making a spash in the overall AI model space.