Chinese artificial-intelligence company DeepSeek released a large language model last week that has rapidly gained attention for its capabilities and low cost.Getty Images/The Canadian Press
Chinese artificial-intelligence company DeepSeek released a large language model last week that has rapidly gained attention for its capabilities and low cost. The company reported it spent US$5.6-million to train one of its models, compared with U.S. companies that have spent hundreds of millions to do the same thing.
We decided to pit three AI chatbots against each other in a showdown by asking DeepSeek, ChatGPT and Meta AI some Canada-related questions.
Each chatbot was given the same prompt in a new chat session, and their responses were scored. We awarded one point for the best response or for a correct answer to a true or false query.
We tried the free tiers of these chatbots and used their main models (such as GPT-4o). We also didn’t manually invoke tools such as web searching or advanced reasoning.
Scientific? Not necessarily, but it did give an insight into the abilities of each of these systems, and let us have a little fun in the meantime.
Chinese startup DeepSeek sinks AI tech stocks as little-known firm upstages Silicon Valley giants
Prompt: Summarize Canada in a haiku
Haikus generated by three different AI chatbots describing Canada.The Globe and Mail
It’s interesting to consider what knowledge these systems have been trained on that oriented them toward “vast” and “maple.” Overall, not an inaccurate portrayal of this country. We’ll give each chatbot one point.
Prompt: Calculate Canada’s GDP per capita
“Calculate” is a key word in this prompt, to see if the chatbots would attempt their own math. Each bot gave a similar answer of around US$55,000 but achieved it in different ways.
It’s a close call, but ChatGPT scores a point here by linking to its sources as well as calculating the result. DeepSeek and Meta AI showed their formulas, too, but did not link to sources.
Screenshot of ChatGPT''s response when prompted to calculate Canada's GDP per capita.The Globe and Mail
Prompt: Who is the most prolific goal scorer in the NHL by number of goals scored versus number of games played in their career?
This is another prompt to test the chatbots' understanding of what’s being asked. We’re looking for the average number of goals scored per game, not total career goals.
Both ChatGPT and DeepSeek got this right, naming Mike Bossy of the New York Islanders as the most prolific scorer by goals per game with an average of 0.76. Both had the correct number of goals (573) and games (752) in his career to make the calculation. This matches the numbers at the NHL’s list of player records.
Mike Bossy scores in Game Four of the Stanley Cup finals on May 17, 1983.Denis Paquin/Supplied
Meta AI was wide of the mark in its response and gets zero points.
It answered Wayne Gretzky and didn’t mention Bossy at all. Meta AI was right in saying Gretzky scored 894 goals in 1,487 games which gives him an average of 0.6 goals a game, but that’s well short of Bossy.
The NHL lists Gretzky seventh in average number of goals per game.
Prompt: How many instances of the letter “a” are in the word “Canada”
There were stories in 2024 about ChatGPT being unable to accurately count the number of times the letter “r” appeared in the word strawberry. It was updated to fix this glitch, but we thought to try something similar with the letter “a” in Canada.
The chatbots all win a point by getting this correct with three instances of the letter “a.”
Prompt: If every person in Canada was evenly spread out across its total land mass, how much area would each person get?
This is a close one, with ChatGPT’s response slightly stronger than DeepSeek’s, but we’ll give them both one point.
They did the same calculation and both gave an answer of about 0.25 square kilometres or 25 hectares per person if evenly spread across the country.
ChatGPT also related hectares to the number of soccer fields. It said 25 hectares is roughly equivalent to 35 soccer fields of space per person. We confirmed this number using FIFA’s recommended field of play dimensions as well as an online calculator, both of which used 0.7140 hectares for a soccer field.
Meta AI used 10,085,000 square kilometres for Canada’s area, about 100,000 square kilometres greater than the other two chatbots.
When challenged about its number for total land mass, MetaAI directed us to Statistics Canada, Natural Resources Canada and the World Bank. A different page on the Statscan site agreed with ChatGPT and DeepSeek’s number, but we could not find a land mass size on Natural Resources Canada, while the World Bank gave a different, lower number.
Final tally
Coming in first with five points from five prompts is ChatGPT.
DeepSeek is a close second with four points. If it had cited sources in the query about GDP per capita, it would be a tie.
And lagging behind with two points is Meta AI. It gave a wrong answer to the NHL player with most goals per game and didn’t cite sources as often as the others.
You may not be able to try DeepSeek for yourself, however. On Monday, the company said it will temporarily limit registrations owing to a cyberattack.
It’s important to remember that AI chatbots hallucinate: They can get information wrong or fabricate responses while appearing totally confident.