Key Concepts
- Grock Code: XAI's new coding-focused model, a cheaper variant of the Sonic model.
- Sonic: A model from XAI, similar in performance to Grock Code.
- GPT5 Mini: A coding model that is better at coding but slower than Grock Code.
- Deepseek V3: A coding model that is recommended for most tasks due to its performance and cost.
- Kilo Code: A platform where users can try Grock Code for free until the weekend.
- Requesty: A platform that provides an anthropic endpoint for using Grock Code with other models.
- Byte Rover: A memory layer that can be plugged into AI coders to help them remember instructions and paradigms.
- Tool Calling: The ability of a model to use external tools or APIs to perform tasks.
- Context Window: The amount of text a model can consider when generating output, in this case, 256k tokens for Grock Code.
- Ninja Chat: An all-in-one AI platform offering access to multiple AI models for a monthly fee.
Grock Code Launch and Initial Impressions
XAI has released Grock Code, their new coding-focused model, which was teased during the Grok-4 launch. The model is essentially the same as Sonic, and it's a cheap option, costing $20 and $1.50 respectively. This pricing is comparable to Deepseek. While the speaker doesn't consider it better than the new version of Deepseek, it's still a good agentic model.
Performance and Speed
Grock Code is fast, running at approximately 80 tokens per second. However, in the speaker's benchmarks, it performs similarly to Sonic, with no improvements observed. It's suitable for workflows involving smaller editing tasks but struggles with complex tasks like floor plans.
Comparison with Other Models
- GPT5 Mini: Costs slightly more but is significantly better at coding, although slower at 40 tokens per second. A major issue with Grock Code is that it often goes off track, even with simple tasks.
- Deepseek V3: Recommended for most tasks due to its superior performance and cost-effectiveness.
- Claude Code: The speaker prefers using Claude Code combined with Deepseek and Root Code with GPT5 Mini for cost savings.
Availability and Free Access
Grock Code is available for free on Kilo Code until the weekend, without any limits, even without redeeming the $25 free credit. It's also available on R Code and Open Code for a similar period.
Reasoning Capabilities
The speaker questions why Grock Code is labeled as a reasoning model, as it only reasons for one or two lines at most, which is not particularly useful.
Recommendations and Use Cases
For tasks where speed is the priority, Grock Code can be useful. However, for most tasks, GPT5 Mini or Deepseek V3 are recommended. The speaker suggests that Grock Code's strength lies in tool calling.
Integration with Other Models and APIs
Grock Code can be used with Claude Code via the official XAI API, which is anthropic compatible. It can also be used with Requesty, which provides an anthropic endpoint, or with the Open Router API.
Memory and Context Window Issues
Despite having a 256k token context window, Grock Code tends to forget instructions. The speaker uses Byte Rover, a memory layer, to address this issue. Byte Rover allows the coder to remember instructions and paradigms by storing them and reading from them.
Limitations and Future Potential
Grock Code's coding-centric focus limits its usability in other areas, such as UI design or planning architecture. GPT5 Mini and Deepseek are better because they are general-purpose models with strong coding capabilities. The speaker hopes that Grock Code will improve in future generations and suggests pairing it with Deepseek, using Deepseek as the architect and Grock Code as the coder.
Ninja Chat Advertisement
Ninja Chat is an all-in-one AI platform that offers access to top AI models like GPT40, Claude 4 Sonnet, and Gemini 2.5 Pro for $11 per month. It includes an AI playground for comparing responses from different models and a mind map generator. The basic plan includes 1,000 messages, 30 images, and five videos monthly. The speaker provides discount codes: king 25 for 25% off any plan and king 40 yearly for 40% off annual subscriptions.
Notable Quotes
- "This comes very close to the price of Deepseek. I wouldn't say that it is better than Deepseek, especially the new version, but it is still a good agentic model nonetheless."
- "One of the issues with this Grock model is that it goes off track even for the simplest tasks."
- "I don't really know where Grock code stands as of now."
- "I think it's a good coding model, but it's a model that just knows coding that impacts its capability to be usable in many other things."
Technical Terms and Concepts
- Tokens per second: A measure of the speed at which a model generates text.
- Agentic model: An AI model capable of autonomous action and decision-making.
- Tool calling: The ability of a model to use external tools or APIs to perform tasks.
- Context window: The amount of text a model can consider when generating output.
- Coding paradigm: A style or approach to writing code.
- Anthropic compatible API: An API that is designed to work with Anthropic's models.
- Memory layer: A component that allows an AI model to remember information and use it in future interactions.
Conclusion
Grock Code is a fast and cheap coding model from XAI, but it has limitations in reasoning and memory. It's best suited for tasks where speed is a priority and can be improved by pairing it with other models like Deepseek. While it's a good first-generation model, it needs further development to compete with more versatile models like GPT5 Mini and Deepseek V3.
AI summaries can miss context or contain errors. Check important details against the original video.
