{"id":111976,"date":"2025-03-04T09:22:33","date_gmt":"2025-03-04T17:22:33","guid":{"rendered":"https:\/\/www.backblaze.com\/blog\/?p=111976"},"modified":"2025-03-04T09:51:07","modified_gmt":"2025-03-04T17:51:07","slug":"experimenting-with-deepseek-backblaze-b2-and-drive-stats","status":"publish","type":"post","link":"https:\/\/www.backblaze.com\/blog\/experimenting-with-deepseek-backblaze-b2-and-drive-stats\/","title":{"rendered":"Experimenting with DeepSeek, Backblaze B2, and Drive Stats"},"content":{"rendered":"\n
\"A<\/figure>\n\n\n\n
<\/div>\n\n\n\n

As we explained in our recent blog post, AI Reasoning Models: OpenAI o3-mini, o1-mini, and DeepSeek R1<\/a>, Chinese startup DeepSeek<\/a> caused a stir when it released its R1 reasoning model in January of this year. Interestingly, DeepSeek R1 has an OpenAI-compatible API, so applications written for OpenAI should<\/em> work with DeepSeek R1 with just a configuration change. Since I had a suitable sample app all ready to go, I decided to put their claim to the test.<\/p>\n\n\n\n

Why, and why not, use DeepSeek?<\/h2>\n\n\n\n

A major difference between DeepSeek and OpenAI is cost. At the time of writing, DeepSeek charges $0.55 per million input tokens and $2.19 per million output tokens for its R1 model. That\u2019s about 3.6% of OpenAI\u2019s $15.00 per million input tokens and $60.00 per million output tokens for its flagship o1<\/code> reasoning model, and about half of o3-mini\u2019s $1.10 per million input tokens and $4.40 per million output tokens.<\/p>\n\n\n\n

Set against this is the fact that, in using the DeepSeek platform\u2019s API, you are sending your data to a startup located in China that has been accused by OpenAI<\/a> of “inappropriately” basing its work on the output of OpenAI’s models. It\u2019s up to you, and your organizations\u2019 data governance policy, whether the trade-off is worthwhile.<\/p>\n\n\n\n

Another consideration is the ability to run DeepSeek\u2019s models locally, on your own infrastructure, or, more likely, your chosen provider\u2019s infrastructure, rather than sending requests to the DeepSeek platform. Spinning up my own DeepSeek instance was out of scope for this blog post, but I\u2019ll likely return to it in a future blog post.<\/p>\n\n\n\n

Swapping OpenAI for DeepSeek<\/h2>\n\n\n\n

Last month, I explained how you can build an AI agent with Backblaze B2, LangChain, and Drive Stats<\/a>, walking you through a simple chatbot that can answer questions based on our Drive Stats<\/a> data set\u201411 years of metrics gathered from the Backblaze B2 Cloud Storage platform\u2019s fleet of hard drives. In that example, the chatbot accepted a natural language question, used OpenAI\u2019s GPT\u20114o mini<\/a> large language model (LLM) to generate a SQL query that might help provide an answer, executed the query against the Drive Stats data set via the Trino SQL engine, and then used OpenAI again to interpret the result set and either repeat the query-interpret cycle, or generate a natural language answer.<\/p>\n\n\n\n

I copied the Jupyter notebook from that example<\/a> and used it as the basis for investigating the feasibility of swapping out OpenAI for DeepSeek. The DeepSeek version of the notebook<\/a> contains the full source code of my experiments; I\u2019ll include relevant extracts here, edited for clarity.<\/p>\n\n\n\n

Since I used the LangChain AI framework, which provides a layer above a range of AI models, the only place that OpenAI surfaced in my code was in creating an instance of LangChain\u2019s ChatOpenAI wrapper:<\/p>\n\n\n\n

# OPENAI_API_KEY must be defined in the .env file
load_dotenv()
llm = ChatOpenAI(model=\"gpt-4o-mini\")<\/pre>\n\n\n\n
The ChatOpenAI class contains all the code required to communicate with OpenAI via its API.<\/p>\n\n\n\n
According to the DeepSeek documentation<\/a>, all you should need to do is:<\/p>\n\n\n\n