Craft offers a variety of AI models to meet different needs and use cases. Below, you’ll find a detailed overview of the AI models available in Craft, along with their key features, best use cases, and parameter counts.
AI Models Available in Craft
| Model | Description | Best For | Parameters | Availability |
|---|---|---|---|---|
| GPT-4o | A highly capable AI model designed to handle complex queries and tasks with precision, leveraging advanced cloud-based processing. | Tasks requiring deep reasoning, creative problem-solving, and high accuracy (e.g., long-form content creation, research, nuanced decision-making). | Not specified; highest-capacity model offering unparalleled capabilities for complex queries. | Cloud-based, requires an internet connection. |
| GPT-4.o-mini | A streamlined version of GPT-4o, optimized for efficiency. | Everyday productivity tasks such as summarizing content, quick Q&A, and general writing assistance. | Smaller version of GPT-4o, balancing speed and resource efficiency while maintaining high quality. | Cloud-based, requires an internet connection. |
| LLaMa 3.2 3B | A multilingual AI model offering advanced on-device performance. | Detailed multilingual tasks like comprehensive translations, content creation in various languages, or handling linguistic diversity. | 3 billion parameters, providing greater capacity and nuance compared to lighter models for on-device use. | On-device, requires download. Data stays on the device and is available offline. |
| LLaMa 3.2 1B | A lightweight, multilingual model focused on conserving device resources. | Quick multilingual tasks such as casual translations or short interactions. | 1 billion parameters, enabling efficient handling of simple tasks with minimal computational power. | On-device, requires download. Data stays on the device and is available offline. |
| DeepSeek R1 7B | An on-device AI model balancing advanced reasoning capabilities with efficiency. | Sophisticated reasoning and logic-based tasks such as structured outlines, solving intricate problems, or creating in-depth analyses offline. | 7 billion parameters, offering robust reasoning capabilities for advanced offline functionality. | On-device, requires download. Data stays on the device and is available offline. |
| DeepSeek R1 1.5B | A lightweight model optimized for fast performance in mathematical and computational tasks. | Math-heavy workflows and lightweight reasoning tasks like solving equations or performing calculations. | 1.5 billion parameters, ensuring efficient task handling while conserving device resources. | On-device, requires download. Data stays on the device and is available offline. |
What Are Model Parameters?
| Parameter Count | Explanation |
|---|---|
| Higher Parameters (e.g., 7B) | Offer sophisticated, nuanced responses, ideal for complex tasks. |
| Lower Parameters (e.g., 1B or 1.5B) | Prioritize speed and efficiency, ideal for simpler or lightweight operations. |
Security Considerations: On-Device vs. Cloud AI
When choosing between on-device models (LLaMa, DeepSeek) and cloud-based models (GPT-4o, GPT-4.o-mini), it's important to consider security implications.
On-Device Models: Privacy & Local Processing
- All data is processed locally, ensuring it stays on your device.
- Ideal for users and organizations prioritizing privacy and security, as no information is transmitted externally.
- Works offline, reducing dependency on an internet connection.
- Since these models run on-device, they are not automatically updated—users may need to manually update them for improvements or security patches.
Cloud-Based Models: OpenAI’s Security & Data Handling
- GPT-4o and GPT-4.o-mini are cloud-based, meaning data is sent to OpenAI’s servers for processing.
- OpenAI implements strict security measures to protect user data, including:
- End-to-end encryption for data transmission.
- Data retention policies ensuring API requests are not used for training unless explicitly permitted.
- Enterprise-grade security compliance, including SOC 2 and ISO 27001 certifications.
- Cloud models allow for frequent updates and improvements, ensuring users always have access to the latest enhancements.
- While cloud models offer higher scalability and accuracy, they require an internet connection, and users should be aware of data-sharing implications.
For more details on how we handle AI security and privacy, check out our Craft AI Assistant Terms.
Which Model Should You Use?
- If privacy and offline access are your priority, choose an on-device model (LLaMa or DeepSeek).
- If you need the most advanced AI capabilities, opt for a cloud-based model (GPT-4o or GPT-4.o-mini), understanding that it requires an internet connection and external data processing.
If you have any questions or need further clarification, feel free to reach out! 🚀