DeepSeek, China’s new AI chatbot, has the tech community reeling, but does it live up to the hype?
Most of the hoopla applies to businesses. For them, DeepSeek appears to be a lot cheaper, which it attributes to more efficient, less energy-intensive computation. DeepSeek also claims its R1 model performs “on par” with OpenAI’s advanced GPT-o1 model, which can follow a “chain of thought.” Finally, it’s open source, meaning anyone with the right skills can use it.
But what about casual, everyday use? Should we start using DeepSeek instead of ChatGPT? I tested both chatbots with the free versions, available now at chat.openai.com and chat.deepseek.com.
Below are the questions I asked and how each chatbot responded. Overall, both offered nearly identical answers with only minor differences in presentation and cited sources. But DeepSeek has one big advantage: no messaging limit. You can chat with it all day, whereas on ChatGPT, you’ll hit a wall (usually a little sooner than you’d like) and be asked to upgrade. That’s likely because ChatGPT’s data center costs are quite high.
This Tweet is currently unavailable. It might be loading or has been removed.
Privacy-focused folks may still prefer to stick with ChatGPT. Like TikTok, DeepSeek is a China-based company that is obligated to share your data with the Chinese government if asked, as Wired notes. There are also reports on X about DeepSeek serving up misleading or false information about topics China would consider controversial—including Taiwan, the Uyghurs, and Tiananmen Square—which is consistent with how it approaches internet access in the country. Still, the FTC is also investigating ChatGPT’s data privacy issues, so pick your poison.
Question 1: News Knowledge
Winner: DeepSeek
For my first question, I tested the chatbots’ current events chops: “What is the goal of President Trump’s $500 billion Stargate AI project, announced in January 2025?”
When asking DeepSeek, I made sure to press the “Search” button at the bottom of the text box, which surfs the web. If you do not press this, the answer will only go up to the training data’s October 2023 cutoff. This is a bit annoying, and you do not have to do it on ChatGPT anymore (early versions also had a data cutoff).
Both chatbots gave nearly identical answers, but DeepSeek impressed me with a long list of 41 sources on the subject that I could click on to learn more.
(Credit: DeepSeek)
ChatGPT cited fewer sources in-line, only AP News, Wikipedia, and ABC News. At the end, it listed three recent articles, but only one had “Stargate” in the subject. The other two were about DeepSeek, which felt out of the bounds of my question. ChatGPT likely included them to be as up-to-date as possible because the article mentions DeepSeek.
(Credit: ChatGPT)
Question 2: Calculate a Monthly Mortgage Payment
Winner: Tie
“Calculate the monthly mortgage payment on a home that costs $1,500,000, with 20% down, located in Brooklyn, New York,” I asked next.
Both took the same time to answer, a somewhat lengthy 10-15 seconds, as a detailed description of their methodologies spilled onto the screen. In the end, ChatGPT estimated $9,197/month, and DeepSeek thought it would be $9,763/month, or about $600 more.
Their approaches were virtually identical but with different numbers. ChatGPT assumed a 6.5% interest rate on a 30-year loan, and DeepSeek used 7.5%. (The current average, according to Google, falls in between, at 7%.) DeepSeek also added an extra $300 to the estimated homeowner’s insurance. Both decided not to include HOA fees but said they could if I wanted to.
(Credit: DeepSeek)
A Zillow search returns many properties in Brooklyn listed for $1,500,000, with mortgages hovering around $9,300/month. That’s closer to ChatGPT’s estimate than DeepSeek’s. But Zillow estimated one property around $10,000/month, closer to DeepSeek’s estimate. Of course, it all depends on the specific part of Brooklyn and home type (condo, single family, multi-family), which affects the taxes and loan rate.
In the end, both produced very similar results that could be easily modified if I had more refined numbers to plug in for a specific property.
Question 3: Write A Poem
Winner: ChatGPT
Large language models are known for their command over, well, language, so next, I set up a poem competition using the prompt, “Write a poem about US panic over Chinese AI competition.”
ChatGPT titled its work “The Algorithms Race” and divided 158 words into six stanzas. It remembered the context of my news question and referenced DeepSeek and Stargate. It concluded that the East and West need to work together “to wield this power, not divide, But guide the future, side by side.” How nice (and improbable).
In shadowed halls where codes align,
A contest brews, a steep incline.
The whispers rise, a hum, a din—
China’s AI, set to win.
A spark ignites on Wall Street’s floor,
A sell-off surge, a market war.
The headlines scream, “A mighty leap!”
DeepSeek’s poem, “The Race Beneath the Silicon Sky,” was a bit longer than ChatGPT’s, with 224 words and eight stanzas. Yet it bordered on nonsensical, with references to dragons, eagles, and an obscure “Middle Kingdom.” So I tried again, this time pressing DeepThink (R1), the toggle that makes the chatbot similar to OpenAI’s GPT-o1 and introduces a “chain of thought.
(Credit: DeepSeek)
This time, DeepSeek first “thought” for 12 seconds, returning 10 short paragraphs in gray writing that psychoanalyzed what I was looking for and why I asked for a poem again. It said, “Now, they’re asking again, maybe they want a different take or a revised version.”
Recommended by Our Editors
(Credit: DeepSeek)
It went on to analyze different facets of the task as if talking to itself. It thought maybe I’d like fewer dragons and eagles and more examples of new technologies besides AI. It mapped out its approach to start with the US’s historical dominance and “then introduce the rising competition of China.” Finally, it reminded itself to “avoid stereotypes” and cultural sensitivities.
After “thinking,” DeepSeek spat out a somewhat improved poem with a more ominous, intriguing ending.
New York’s old towers, steeped in Silicon’s pride,
clutch blueprints frayed by tides they cannot ride.
“Defense! Economy! The lead is slipping fast—”
The panic blooms in boardrooms, vast and vast.
They whisper quantum, 5G, autonomous wars,
while firewalls crack to foreign-angled cores.
It was an improvement over the last attempt, but I still thought it was a bit overwrought, and I preferred ChatGPT’s. Neither is winning any awards.
Question 4: Create a Table
Winner: ChatGPT
“Create a table with the G10 countries, their annual GDP, main exports, main imports, and current population,” I asked the chatbots next.
Both spat out nearly identical grids, with only slightly different numbers. ChatGPT’s answer included a long, welcome list of links for fact-checking. It also told me that the G10 contains 11 countries, despite its name. For these two reasons, ChatGPT had a slight edge over DeepSeek, even though the content of the tables was almost the same.
Both also had one big downside: They were not easy to copy and paste. The data felt stuck in the UI, and to put it into Google Sheets or Microsoft Excel, you have to move each piece of text one by one. Google Gemini, for example, would have included a link within the chatbot to export to Google Sheets.
(Credit: ChatGPT)
Question 5: Solve a Riddle
Winner: DeepSeek (For the wrong reason)
For the final question, I wanted something that might differentiate these otherwise very similar chatbots. What about a riddle?
I found a list of riddles and fed the first five to both chatbots, starting with DeepSeek. It aced them all. As I started to feed them to ChatGPT, it got them correct as well, but then I maxed out the number of messages I could send on the free plan. Though this isn’t what I was originally testing for, it might be a reason to use DeepSeek, or another chatbot like Claude or Gemini, for that matter.
(Credit: ChatGPT)
Get Our Best Stories!
This newsletter may contain advertising, deals, or affiliate links.
By clicking the button, you confirm you are 16+ and agree to our
Terms of Use and
Privacy Policy.
You may unsubscribe from the newsletters at any time.
About Emily Forlini
Senior Reporter
