# Grok 4 Fast doesn't make any sense

Source: https://www.youtube.com/watch?v=PVhVq9RDxwM
Recap page: https://rapidrecap.app/video/PVhVq9RDxwM
Generated: 2025-09-21T05:32:13.558+00:00

---
## Quick Overview

Grok 4 Fast, a new multimodal reasoning model from xAI, demonstrates remarkable performance and cost-efficiency, achieving top rankings on various benchmarks while being significantly cheaper than competitors like GPT-4 and Claude.

**Key Points:**
- Grok 4 Fast achieves state-of-the-art performance with a superior price-to-intelligence ratio compared to other publicly available models.
- It outperforms Grok 3 Mini across reasoning benchmarks and slashes token costs.
- The model demonstrates comparable performance to Grok 4 on benchmarks while using 40% fewer tokens on average.
- Grok 4 Fast ranks highly on the ARC-AGI leaderboard, showing strong performance in reasoning and various benchmarks.
- The model is available for free on Grok.com, Grok.x.com, iOS and Android apps, and OpenRouter.
- xAI is making new bets on scaling RL & post-training, focusing on data, deep thinking, and a specific training recipe.
- The company is actively recruiting talent to harness the power of their Colossus 2 infrastructure.

![Screenshot at 00:01: The title slide for the video, 'Grok 4 Fast: Pushing the Frontier of Cost-Efficient Intelligence', is displayed, featuring a striking image of a hummingbird superimposed over digital data streams, symbolizing speed and intelligence.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-00-01.png)

**Context:** The video discusses the release and capabilities of Grok 4 Fast, a new AI model developed by xAI. It highlights the model's performance, cost-efficiency, and its positioning within the competitive landscape of large language models. The presentation includes comparisons with other leading models and references benchmark results, emphasizing the significance of Grok 4 Fast's achievements.

## Detailed Analysis

Grok 4 Fast represents a significant advancement in AI, particularly in cost-efficient intelligence. Developed by xAI, it achieves state-of-the-art performance, outperforming its predecessor, Grok 3 Mini, in reasoning benchmarks while substantially reducing costs. An independent review from Artificial Analysis confirms Grok 4 Fast's superior price-to-intelligence ratio, positioning it favorably against competitors like GPT-4 and Claude models. The model is accessible through various platforms, including Grok.com, Grok.x.com, and mobile apps. xAI's strategy involves scaling Reinforcement Learning (RL) and post-training, focusing on data, deep thinking, and a refined training recipe. The company's rapid progress is attributed to its agile team and its investment in powerful infrastructure like Colossus 2. The model's performance is further validated by its high rankings on benchmarks like ARC-AGI, where it excels in reasoning and general capabilities, even surpassing models like GPT-4 in some aspects while being significantly cheaper. The development team is actively seeking talent to further enhance these models and leverage their advanced infrastructure.

### Grok 4 Fast Performance

- Achieves SOTA results with superior price-to-intelligence ratio
- Outperforms Grok 3 Mini in reasoning benchmarks
- Comparable performance to Grok 4 with 40% fewer tokens

### Accessibility

- Available for free on Grok.com, Grok.x.com, iOS, Android, and OpenRouter
- General availability via xAI API with competitive pricing

### xAI Strategy

- Focus on scaling RL & post-training
- Bets on data, deep thinking, and specialized training recipes
- Emphasis on team speed and leveraging Colossus 2 infrastructure

### Benchmark Performance

- Top rankings on ARC-AGI leaderboard
- Excels in reasoning and general capabilities
- Outperforms competitors on cost-effectiveness

### Competitive Landscape

- Significantly cheaper than GPT-4 and Claude
- Competitive with Gemini models on Arena Score and cost

### Future Outlook

- Seeking talent to harness Colossus 2
- Continued development to improve AI capabilities and efficiency

![Screenshot at 00:00: The title slide of the video, 'Grok 4 Fast: Pushing the Frontier of Cost-Efficient Intelligence', is displayed with a hummingbird image symbolizing speed and intelligence.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-00-00.png)
![Screenshot at 00:10: A scatter plot titled 'Intelligence vs. Price' shows various AI models plotted by their Arena Score against their cost per intelligence index. Grok 4 Fast is positioned favorably, indicating high intelligence at a low cost, marked as '47x cheaper' than other models.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-00-10.png)
![Screenshot at 02:04: An ARC-AGI-2 Leaderboard chart displays models ranked by score against cost per task, highlighting Grok 4 \(Thinking\) at the top with a high score and relatively low cost.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-02-04.png)
![Screenshot at 02:35: A tweet from Dustin Tran details his departure from Google DeepMind and his work on Gemini, including leading post-training evaluations and innovations in reasoning.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-02-35.png)
![Screenshot at 03:16: A tweet from xAI announces Grok 4 Fast, a multimodal reasoning model with a 2M context window, setting a new standard for cost-efficient intelligence.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-03-16.png)
![Screenshot at 03:25: A table compares the performance of Grok 4 Fast against Grok 4, Grok 3 Mini, GPT-5, and GPT-5 Mini across several benchmark tests, showing Grok 4 Fast's competitive scores.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-03-25.png)
![Screenshot at 03:50: The Search Arena leaderboard displays rankings of LLMs based on community votes, with Grok-4-fast-search at the top, followed by O3-search and Gemini 2.5 Pro.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-03-50.png)
![Screenshot at 06:02: A slide illustrating 'OpenAI: Scaling Reinforcement Learning' shows a progression of models \(GPT-4o, 01, 03, GPT-n, Next\) with increasing RL Compute relative to Pretraining Compute.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-06-02.png)
![Screenshot at 10:03: The LM Arena leaderboard ranks various models based on their performance across different categories, with Grok-4-fast showing strong results.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-10-03.png)
![Screenshot at 13:41: A bar chart from 'AI Training Datacenter Capacity by Frontier AI Lab' shows projected compute capacity, with xAI's blue bars indicating significant growth over time.](https://ss.rapidrecap.app/screens/PVhVq9RDxwM/00-13-41.png)