Building an AI Agent with DeepSeek Using StackAI

With the rise of AI agents, the future of work is changing rapidly. AI agents are performing jobs that have traditionally required human intelligence, such as IT support, investment analysis, and physician assistants. The introduction of large language models has made this transition possible.

One of the most hyped LLMs on the market today is DeepSeek. Founded by a Chinese firm in 2023, DeepSeek significantly reduces computational costs while maintaining high performance. The latest model of DeepSeek — R1 — competes directly with models like OpenAI’s o1 in areas such as advanced reasoning, math problem-solving, and code generation.

Stack AI was one of the first platforms to offer support for DeepSeek. Stack AI allows enterprise teams to build AI agents using a drag-and-drop interface. Customers of Stack AI leverage DeepSeek in AI agents for real-world use cases.

In the following blog, we’ll give an overview of DeepSeek, including cost, performance, and accuracy of the model. We’ll also show you how to build an AI agent powered by DeepSeek using Stack AI.

Using DeepSeek LLM in StackAI

Let’s connect DeepSeek to Stack AI. Navigate to the account dashboard and click ‘New Project’.

Click the ‘Quick Start’ option. Select TogetherAI or Groq as providers, and look for the available DeepSeek models.

Customize your AI agent with a few clicks. Whenever ready, click the Deploy button.

This is one of the simplest examples of how to use the DeepSeek LLM in Stack AI. But you can leverage DeepSeek in any of your AI Agent in Stack AI, from the most complex, to the most facile.

For more complex use cases, choose the ‘Workflow Builder’ option. This will present you with the drag-and-drop interface. Click the LLMs tab on the left side of the screen, and select the DeepSeek model you want to use.

Understanding DeepSeek LLM

DeepSeek is an advanced open-source AI model. Launched in January 2025, this Chinese-developed chatbot is designed to handle a wide range of tasks, including natural language processing, code generation, and complex reasoning.

DeepSeek employs Mixture of Experts (MoE), which sets it apart from other AI models like GPT-4 or Llama 3. This approach results in faster processing while maintaining higher accuracy.

DeepSeek utilizes several original innovations, such as GPRO, FP8 mixed precision training, and DualPipe. DeepSeek’s performance achieves parity with models like GPT-o1, while requiring many fewer GPUs to operate.

DeepSeek is open source, enabling global developers and organizations to access, modify, and implement the technology. DeepSeek's speed is a standout feature, thanks to several key innovations:

1. FP8 Precision: By using 8-bit floating point instead of the traditional 32-bit, DeepSeek reduces memory usage by 75% while maintaining accuracy, enabling faster training and inference speeds.
2. Multi-Token Prediction: Unlike traditional models that generate responses one token at a time, DeepSeek predicts multiple tokens in parallel, significantly improving processing speed.
3. Mixture of Experts (MoE) Framework: This approach activates only a fraction of the model's 671 billion parameters for each request, leading to lower computational costs and more efficient scaling.
4. Optimized Hardware Usage: DeepSeek employs techniques like PTX programming and the DualPipe algorithm to maximize GPU efficiency and improve communication between GPUs during training.

DeepSeek has demonstrated competitive accuracy across various domains. DeepSeek achieved a 90% accuracy rate on advanced benchmarks, surpassing GPT-4o and outperforming ChatGPT's 83% accuracy rate. In the HumanEval coding tests, DeepSeek delivered a 73.78% pass rate, showcasing its proficiency in code generation and debugging.

DeepSeek's innovative approach to AI development has positioned it as a significant player in the global AI landscape. Its ability to match the performance of leading US models at a fraction of the cost has immediately turned it into a top contender in the LLM space.

DeepSeek vs. ChatGPT: Key Differences and Comparison

ChatGPT shocked the world in 2022, and it’s still the go-to comparison for other chatbots. So let’s put ChatGPT to DeepSeek side-by-side. While both offer impressive capabilities, they have distinct strengths and approaches that set them apart.

ChatGPT uses a traditional transformer model, which processes all parameters for every task. In contrast, DeepSeek employs a Mixture-of-Experts (MoE) approach, activating only relevant subsets of its 671 billion parameters for each query. This allows DeepSeek to be more adaptable to different types of tasks.

DeepSeek excels in technical tasks, particularly in mathematics, where it achieves a 90% accuracy rate, compared to ChatGPT's 83%. ChatGPT, however, demonstrates stronger capabilities in understanding context and providing nuanced responses across a broader range of topics.

DeepSeek takes an open-source approach, making it freely available and customizable. You can run the model on local machines, thanks to its smaller footprint. ChatGPT operates on a freemium model, offering basic features for free but requiring a subscription for advanced capabilities. For specific pricing, let’s compare the pricing of DeepSeek’s R1 model versus OpenAI’s o3-mini.

OpenAI o3‑mini costs approximately $1.10 per million input tokens and $4.40 per million output tokens. DeepSeek R1, on the other hand, costs around $0.14 per million input tokens (if using cache hits) and about $2.19 per million output tokens, making R1 cheaper overall on a per-token basis.

Both models perform well in various scenarios, but with different strengths:

• Content Creation: DeepSeek effectively organizes information and clarifies reasoning, while ChatGPT offers structured layouts with main points.
• Coding: ChatGPT often delivers correct code immediately, while DeepSeek may require adjustments but provides a more engaging interface.
• Summarization: Both models excel, with DeepSeek quickly highlighting key details and ChatGPT providing comprehensive breakdowns.

To summarize, ChatGPT is ideal for general-purpose tasks and user-friendly interactions. DeepSeek focuses on specialized technical applications, and offers greater customization as an open source model.

Why Use Stack AI for AI Agent Development

Stack AI empowers enterprise teams to build AI agents powered by DeepSeek. With a drag-and-drop builder interface, Stack AI allows non-technical teams to create agents for specific use cases. These agents solve key business problems in companies including finance, education, and healthcare.

Stack AI offers flexibility in choosing language models from major providers like OpenAI, Anthropic, Google, and Meta, allowing users to select the most suitable model for specific tasks. For more specialized applications, the platform integrates with various AI infrastructures, including:

• Hugging Face for community models
• Cerebras for rapid inference and training
• Azure AI for secure, HIPAA-compliant OpenAI model deployment

Stack AI integrates with numerous 3rd party platforms, such as Microsoft Sharepoint, AWS S3, Salesforce, and HubSpot. Users can connect and organize their data sources into knowledge bases.

To streamline development, Stack AI provides ready-to-use templates for common applications, including chatbots, RFP response agents, and contract analyzers. See Stack AI’s Template Gallery for a full overview.

The platform prioritizes data protection and adheres to industry-standard compliance regulations like HIPAA, GDPR, and SOC, ensuring privacy and security for various sectors. This allows enterprise teams to deploy their AI agents in daily operations without requiring additional compliance solutions.

Stack AI: The Best AI Builder Tool for DeepSeek

DeepSeek has emerged as one of the most formidable LLMs on the market. As an open source model with high performance and low cost, it now rivals the most established players on the market, including Open AI.

Stack AI is the only AI builder tool that allows you to create AI agents powered by DeepSeek, while still maintaining all the enterprise capabilities teams at large companies need.

If you want to build enterprise AI agents for your use case, try Stack AI for free to create the tools your team needs to succeed.