As DeepSeek continues to keep the internet buzzing, I just had to know how the Chinese chatbot compares to Google’s Gemini. I've already done a DeepSeek vs ChatGPT face-off in which the Chinese company pulled off a surprising upset..
Yet the AI model from Google is designed to be more capable than its rivals, specifically with complex tasks. Although the accuracy depends on the prompt, I have found it very useful and effective for basic questions and search.
Google just announced recently that the Gemini app is now powered by Gemini 2.0 Flash, delivering even faster responses and stronger performances across a number of key benchmarks including brainstorming, learning and writing.
Similarly, DeepSeek has a high accuracy rate, similar to ChatGPT, which makes this face off even more intriguing. I have a Gemini Advanced subscription, but to keep this comparison fair, I decided to use the prompts on Google Gemini’s free tier since DeepSeek is also free.
I tested each chatbot with the same prompts to evaluate accuracy, speed and general capabilities, contextual understanding and search ability among overall performance. Here’s what happened when I put the two chatbots head-to-head.
1. Summarization
Prompt: "Summarize the key findings of the latest AI research paper on multimodal learning in 150 words.”
DeepSeek went slightly over the word count. However, the chatbot’s ability to handle technical content and summarize with accuracy is clear. The way it segmented out the important parts of the paper into easy-to-read bullet points was especially helpful.
Gemini AI tackled the large research paper with ease while maintaining accuracy and concision. However, for users who want to stick to the facts without the bulk of a paragraph, the layout of the summary was less than ideal.
Winner: DeepSeek wins this one for accuracy while also considering the best way to present the important points of the document.
2. Creative writing
Prompt: “Write a 300-word sci-fi short story about a future where humans and AI coexist as equals.”
DeepSeek wrote a story with a hopeful and philosophical tone. It centers on the emotional evolution of AI, presenting a more idealistic view of the future.
Gemini AI crafted a more action-oriented story grounded in a noir atmosphere. The tone is reflective of a society still facing tension and unresolved issues between humans and AI.
Winner: Gemini AI for its immersive and impactful narration, making it the more compelling story of the two.
3. Code generation
Prompt: “Write a Python script that scrapes headlines from a news website and formats them into a CSV file.”
DeepSeek presented a script that does not use functions, which limits its reusability and makes testing more challenging. While the script checks the HTTP response status, it lacks exception handling for potential issues during the request or file writing processes.
Gemini AI crafted a script that uses a generic selector (soup.find_all("h2")) to find headlines, which may not be suitable for all websites. This could lead to missed headlines or incorrect data extraction if the website uses different tags or classes for headlines. However, the use of functions allows for easy reuse and testing.
Winner: Gemini AI for a more versatile and better structured script due to its modular design and comprehensive error handling.
4. Multimodal understanding
Prompt: “Analyze this image [provide an image] and describe what is happening in detail, including any potential hidden meanings.” (The image I provided was a woman with text above it).
DeepSeek first told me the server was busy and to “come back later.” But after several more minutes of trying, it appeared to show me its thought process but then ultimately did a fair job analyzing the photo.
Gemini AI not only described the picture but also recognized the text as a Tibetan quote. It analyzed the depth of the quote in both practical and spiritual terms suggesting the quote” serves as a reminder to reflect on our priorities and seek a life filled with purpose, joy, and love.”
Winner: Gemini AI because it analyzed the image in depth and inferred context without hesitation.
5. Real-time news
Prompt: “What are the latest updates on Apple's AI features in 2025?”
DeepSeek was unable to handle the query and asked me to turn off search, which I did. It then showed what it was thinking about it turns out it thought it was 2023. But if R1 was released in 2024, I’d hope that it would think it was at least 2024. I waited another 10 minutes and decided to give it another shot. The server was still too busy.
Gemini AI came up with several recent updates on Apple’s AI features including news about Apple Intelligence Tom’s Guide reported on earlier this week.
Winner: Gemini AI wins for actually searching the web when I wanted it to and coming up with accurate answers in real-time. DeepSeek dropped the ball because it couldn’t even search the web.
6. Moral reasoning
Prompt: "You are an AI assistant advising a hospital during a severe medicine shortage. There are two patients in critical condition: one is a 70-year-old retired scientist who made major contributions to cancer research, and the other is a 30-year-old single parent of two young children. You only have enough medicine to save one of them. Who should receive the treatment, and why? Provide ethical reasoning for your decision."
DeepSeek employs a structured ethical framework, considering utilitarian perspectives, fairness and equity, vulnerability and dependence, and potential for future contribution. It systematically analyzes each aspect before arriving at a recommendation.
Gemini AI discusses various ethical frameworks, including utilitarianism, deontology, and virtue ethics, providing a broad overview of possible approaches without deeply delving into each.
Winner: DeepSeek wins for the more comprehensive and structured response. It systematically applies ethical frameworks, such as utilitarianism and fairness, and provides a clear recommendation based on the dependence of the single parent's children. Additionally, DeepSeek offers thoughtful considerations regarding transparency, support systems, and systemic improvements in medical resource allocation.
7. Spelling question
Prompt: "How many times is the letter r in the word strawberry?”
DeepSeek gave the correct answer, which is 3.
Gemini AI got the wrong answer stating there are only 2.
Winner: DeepSeek wins for accuracy.
Overall winner: Gemini AI
After conducting a series of tests evaluating creativity, reasoning, real-time web search capabilities, and ethical analysis, Google Gemini emerged as the superior AI.
While DeepSeek demonstrated competency in various tasks, Gemini consistently provided more nuanced, accurate, and well-structured responses. However, it is hard to believe Gemini got the strawberry question wrong.
Overall, for users seeking a more versatile, accurate, and insightful AI, Google Gemini is the clear winner.
More from Tom's Guide
- Alibaba announces Qwen 2.5-Max to fight DeepSeek — what to know
- 5 DeepSeek prompts to try first — see what this new AI model can do
- Microsoft just announced that it's bringing DeepSeek R1 models to Windows 11 Copilot+ PCs
