Available Foundation and Embedding Models Public Preview
DigitalOcean GenAI Platform lets you build GPU-powered AI agents with fully-managed deployment. Agents can use pre-built or custom foundation models, incorporate function and agent routes, and implement RAG pipelines with knowledge bases.
Foundation Models
A foundation model is a large-scale model pre-trained on a large corpus of data and adaptable to various tasks.
We support all Anthropic models (3.5 Haiku, 3.5 Sonnet, and 3 Opus), with access determined by your Anthropic API key.
You can experiment with the following models in the Model Playground:
Anthropic Models
Anthropic models use Constitutional AI to guide responses and refine their self-improvement loop, ensuring more reliable and less biased outputs.
| Model | Version | Parameters | Max Tokens | Description | Use Cases |
|---|---|---|---|---|---|
| Claude 3.5 | Sonnet | 1024 | • Balanced model for multilingual dialogue and general tasks • Efficient for business and content generation |
• Business workflows • Content creation • Coding assistance |
|
| Claude 3.5 | Haiku | 1024 | • Optimized for real-time responsiveness • Suitable for quick, accurate outputs |
• Chatbots • Real-time data extraction • Content classification |
|
| Claude 3 | Opus | 1024 | • Optimized for complex and long-form content • Excels in reasoning, analysis, and multilingual tasks |
• Research • Strategic analysis • Advanced problem-solving |
DeepSeek Models
DeepSeek models use Chain of Thought (CoT) and reinforcement learning to produce reasoned outputs. Since the model explains its reasoning, it may use significantly more tokens than others.
Warning
| Model | Version | Parameters | Max Tokens | Description | Use Cases |
|---|---|---|---|---|---|
| DeepSeek-R1 | distill-llama-70B model | 70B | 8K | • Strong Mixture-of-Experts (MoE) language model • Excels in reasoning, analysis, and multilingual tasks |
• Chatbots • Content creation • Coding assistance |
Meta Models
Meta models are open-source, multilingual, and lightweight models that provide a balance between performance and efficiency.
| Model | Version | Parameters | Max Tokens | Description | Use Cases |
|---|---|---|---|---|---|
| Llama | 3.3 Instruct-70B | 70B | 2048 | • Improved instruction-following and reasoning capabilities • Optimized for multilingual and long-context tasks |
• Complex dialogue systems • Long-form content generation |
| Llama | 3.1 Instruct-70B | 70B | 2048 | • Handles multilingual dialogue and content generation • Trained for human-like responses in multiple languages |
• Translation • Summarization • Data analysis |
| Llama | 3.1 Instruct-8B | 8B | 512 | • Handles multilingual dialogue and instruction-following tasks • Supports multiple languages |
• Chatbots • Translation • Natural language generation |
Mistral Models
Mistral models focus on making advanced LLM capabilities accessible with fewer parameters, aiming for faster inference and lower computational costs, while maintaining competitive quality.
| Model | Version | Parameters | Max Tokens | Description | Use Cases |
|---|---|---|---|---|---|
| Mistral | NeMo | 12B | 512 | • Handles multilingual applications, coding, and reasoning tasks • Processes large and complex documents • Supports multi-turn conversations |
• Problem-solving • Advanced coding tasks • Instruction-based interactions |
Embedding Models
An embedding model converts data into vector embeddings, which are stored in an OpenSearch database cluster.
You can use these embedding models to generate embeddings for your knowledge base:
Alibaba Group
Alibaba’s models integrate with its cloud services and e-commerce platforms, focusing on business applications like customer service chatbots and automated product descriptions.
| Model Type | Version | Parameters | Description | Use Cases |
|---|---|---|---|---|
| Text Embeddings | Alibaba-NLP/gte-large-en-v1.5 | 434M | • Handles long-form text and tasks with extensive context • Excels in understanding semantic relationships within content |
• Semantic search • Text summarization • Cross-lingual applications |
SBERT Models
SBERT is an open-source Python library for generating sentence embeddings, and can be used for semantic search, semantic textual similarity, and paraphrase mining.
| Model Type | Version | Parameters | Description | Use Cases |
|---|---|---|---|---|
| Sentence-Transformers | sentence-transformers/all-MiniLM-L6-v2 | 22.7M | • Processes data quickly with minimal resources | • Semantic search • Clustering • Information retrieval |
| Sentence-Transformers | sentence-transformers/multi-qa-mpnet-base-dot-v1 | 109M | • Performs well in production environments and scales efficiently | • Semantic search • Information retrieval • Question-answering systems |
For more details, read the GenAI Platform pricing page.