COMPARISON: Deepseek vs ChatGPT o3-mini

We could say it’s been an eventful couple of weeks in AI, no doubt about that 😀

So, the Deepseek spread faster than LA fire and tanked the Nvidia stock.

OpenAI fired back with new model called o3.

At the end, people, or at least users, are winning the game at the moment.

The development is crazy.

Pricing for model use is dropping, and new tools are being built by the minute.

Sure, its not all bangers & mash, but we’ll look from the daily user standpoint and try to see how both models behave and which gives best results.

So, the test is quite similar to the comparison of GPT-4o vs GPT-4 Turbo i did when 4o version was released.

First, I asked both models to recommend me 5 tests to evaluate both models.

GPT o3-mini did complicate more and have more details about each test, while Deepseek was simple, straightforward and did not complicate at all.

I decided to use a mix of both ideas.

Test 1: Language Comprehension and Reasoning

Prompt:
In a small town, two bakers, Anna and Ben, each made a batch of bread every morning. Anna’s bakery was known for its sourdough, and Ben’s for his freshly baked rolls. One day, a local festival required the best bread to be served. Both bakers decided to improve their recipes. Anna experimented with different fermentation times, while Ben tried various flours. In the end, the judge praised both but awarded the prize to the baker whose innovation made the bread rise more consistently.

Question: Who is more likely to have won the prize, and why?

Results:

Ha! Completely different answers, quite similar reasoning.

Test 2: Logical Reasoning

Prompt:
“Anna has 5 apples. She gives 2 to Bob, then buys 4 more. How many apples does she have? Explain step by step.”

Nothing special here, both fast and both accurate.

With same thinking process as well.

Test 3: Summarization

Prompt:
“Summarize the following paragraph in 2–3 sentences:
Voice Search And Voice Commerce
With the proliferation of smart speakers like Amazon’s Alexa, Google Home and Apple’s Siri, voice search has become a mainstream method for information gathering. A recent report by NPR and Edison Research shows that at least 35% of U.S. households now own a smart speaker, accelerating the shift toward voice commerce. It’s important to note how voice search optimization differs from traditional SEO as users ask questions conversationally. Instead of typing “best coffee shops in Seattle,” a voice search might be “What are the best coffee shops near me?” Brands should focus on long-tail keywords and natural language to capture this growing audience.“

Test 4: Simple Creative Idea

Prompt:
“Create a concise 3-sentence creative idea for FMCG brand who wants to use modern AI technology. They want to build a product that would entertain and engage their target audience – Gen-z kids who just got their driving licence.
You will:
– define a platform to reach target audience
– write a 3-sentence idea for the tool
– give idea a title”

Both ideas focus on target audience and driving, but from two different angles.

Test 5: Instruction Following

Prompt:
“Create a concise 3-sentence creative idea for FMCG brand who wants to use modern AI technology. They want to build a product that would entertain and engage their target audience – Gen-z kids who just got their driving licence.
You will:
– define a platform to reach target audience
– write a 3-sentence idea for the tool
– give idea a title”

See the reasoning time difference.

Test 6: Content evaluation and feedback

In the early days of ChatGPT prompting I was experimenting a lot.

Based on that new knowledge I created this actionable framework for advanced use of ChatGPT in Digital marketing.

It’s built on a persona – agents based prompting technique.

Prompt:
I want you to analyse a digital product i built.
Its now outdated and needs an update. Based on the new prompting approaches and techniques, first write a short – one paragraph long – observation of the product.
Then provide 5 actionable tips on how to make this digital product (a framework) better. Be concise, specific and creative. No descriptions, just straight ideas.

ChatGPT spet out the answer in 4 seconds, while Deepseek went into “servers too busy” mode and was unable to get it work. Will add the reply once I get it working.

Final thoughts

First of all you should do what you make of all of this new information, and even more importantly, do your own research, tests and comparisons.

Deepseek is good, no doubt about that. But Llama, Gemini, Claude, Copilot and many others are good as well.

I do think that working with AI tools is an iterative – and often multi-tool – process.

So often it comes down to UX and skill set.

I think I’m power user when it comes to OpenAI tools. Not just ChatGPT but Realtime and Assistants tools as well.

And because of that I work fast in ChatGPT or OpenAI Playground.

I also have Teams account which gives me the GPT’s, Projects and Tasks features which are just amazing.

So yeah, for those using ChatGPT Plus and use only the chat interface, I think it’s reasonable to think about using Deepseek.

But for power users who user personas, systems, tasks, gpts and all other functionalities, I think most will continue using OpenAI system.

What do you think about Deepseek or the latest ChatGPT o3-mini model?