How DeepSeek’s budget-friendly AI mannequin stacks up in opposition to ChatGPT, Claude, and Gemini in SQL, EDA, and machine studying
Nvidia inventory worth slumped over 15% on Monday, Jan twenty seventh, after a Chinese language startup, DeepSeek, launched its new AI mannequin. The mannequin efficiency is on par with ChatGPT, Llama, and Claude however at a fraction of the fee. In line with Wired, OpenAI spent greater than USD$100m to coach GPT-4. However DeepSeek’s V3 mannequin was skilled for simply $5.6m. This value effectivity can be mirrored within the API prices — for each 1M tokens, the deepseek-chat mannequin (V3) prices $0.14, and the deepseek-reasoner mannequin (R1) prices solely $0.55 (DeepSeek API Pricing). In the meantime, gpt-4o API prices $2.50 / 1M enter tokens, and o1 API prices $15.00 / 1M enter tokens (OpenAI API Pricing).
At all times intrigued by rising LLMs and their software in knowledge science, I made a decision to place DeepSeek to the take a look at. My aim was to see how nicely its chatbot (V3) mannequin might help and even substitute knowledge scientists of their every day duties. I used the identical standards from my earlier article sequence, the place I evaluated the efficiency of ChatGPT-4o vs. Claude 3.5 Sonnet vs. Gemini Superior on SQL queries, Exploratory Information Evaluation (EDA), and Machine Studying (ML).
