Dhruv Bhutani / Android Authority
OpenAI’s ChatGPT has dominated the AI chatbot conversation since its 2022 debut. However, if you follow the world of AI, you’d have come across the name Deepseek thrown around over the last few weeks. The Chinese large language model claims to trade blows with ChatGPT for its speed, accuracy, and, most importantly, open-source nature. But what’s truly astonishing is the training efficiency of R1. Relying on pure reinforcement learning versus GPT-4‘s supervised fine-tuning, the entire model cost just $12 million in training versus the $500 million required for the upcoming GPT-5.
Of course, none of that really matters to the end consumer. What matters is if it’s any good in its intended purpose. I’ve spent the last couple of days testing out Deepseek R1 as part of my workflow — ideating, coding, performing tasks like grammar checks, and more. My takeaway? OpenAI needs to be seriously worried.
Rational thinking: A human-like approach
Dhruv Bhutani / Android Authority
Deepseek’s biggest differentiator is its human-like train of thought. Unlike ChatGPT, which presents just the final output of your query, Deepseek R1 shows you how it is coming to the answer. That remarkably human-like internal monologue is how the LLM can present answers that match the human thought process.
Instead of running benchmarks, or traditional ways to gauge the performance of a LLM, I decided to put Deepseek R1 through my daily routine.
Deepseek’s internal dialogue talks you through its reasoning process. The first time I saw this, I was blown away.
As a journalist, I’m vehemently opposed to AI-based writing. However, the tool can be a valuable addition to your arsenal for brainstorming, ideating on talking points, and pointing out loopholes. For the sake of testing, I asked both Deepseek and ChatGPT to craft an outline for an article comparing two different LLM models. My exact query was, “I’m writing an article comparing two different LLM models from an everyday user’s perspective. What talking points should I include? Help me craft an outline.”
The results were astounding. While the two models touched upon the same essential pointers, Deepseek went a step further and broke it down into a structure that was remarkably similar to how I would approach the topic. Moreover, it included very relevant points like ethical considerations for bias, fairness, and transparency that ChatGPT entirely skipped.
Deepseek includes the logical thinking process it went through while coming to the solution, and trust me, the first time I saw this, I was blown away. We might be far away from artificial general intelligence, but watching a computer think like this shows you just how far we’ve come. The internal dialogue is interesting enough that I’m including a brief transcript below.
“Okay, so I need to help someone create an outline for an article comparing two LLM models from an everyday user’s perspective. Let’s start by breaking down what they’re asking for. They want talking points that an average user would care about, not too technical.
First, the introduction should set the stage. Maybe explain what LLMs are in simple terms and why comparing them matters for regular users. Then, the main sections. User experience is probably a big one. That could include how easy each model is to access — like are they free, paid, or freemium? Also, the interface design. Do they have a user-friendly app or website? Maybe compare sign-up processes.”
Next, I put it up to a coding task. LLMs like ChatGPT and Claude might not be capable of full-fledged coding yet, but they can be useful tools to learn how to code. For this task, I gave both Deepseek and ChatGPT the same prompt – “I’m new to programming. Help me write a game of Tic Tac Toe. It should run in pyscript.” Once again, the difference in output was stark.
Dhruv Bhutani / Android Authority
ChatGPT presented me with a code solution, a brief breakdown of the languages it used, and how to run the game. Meanwhile, Deepseek ran me through its entire thinking process of what components were needed to create the game — for example, a game board display, handling user clicks, alternating turns between X and O, and more.
Next, it broke down the HTML structure for drawing interface elements and the Python logic for the game. It also validated its choices and made styling considerations like centering the text. It then detailed not just the features of the game but also how to run it and how to modify it further. This is invaluable information for someone new to coding, and ChatGPT’s response simply doesn’t compare.
Dhruv Bhutani / Android Authority
Screenshot
Alright, back to writing tasks. For this one, I wanted to test out the built-in web search functionality in both LLMs. So, I asked both Deepseek and ChatGPT to write a review of the OnePlus 13. I picked this specific phone because it was past the knowledge update date of both LLMs, and, well, I had the phone in hand to validate the output. While neither LLM is going to take my job any time soon, this is another example where Deepseek’s output was leaps and bounds ahead of ChatGPT.
When ChatGPT presented a review structure, it merely focussed on the specs without adding much explanation and no context whatsoever. Deepseek, on the other hand, drew comparisons with the competition and even highlighted areas where the OnePlus 13 was lacking. As someone who does have the phone in hand, Deepseek’s observations, obviously drawn from existing reviews, were accurate and well-placed.
Deepseek vs ChatGPT: Which one should you pick?
Dhruv Bhutani / Android Authority
After only a couple of days of use, I’m convinced that Deepseek is an excellent alternative to ChatGPT for more reasons than one. Sure, in my tests, Deepseek consistently won in terms of the quality of output — both in terms of context and understanding, but also the explanation of its reasoning. However, with a bit of fine-tuning, ChatGPT can also give similar results. That said, Deepseek has other things going its way, too.
For one, using Deepseek is by and large free for end-customers right now, compared to the rather expensive $20 a month that ChatGPT charges for its higher-end models. That’s a big plus. Moreover, Deepseek’s open-source nature means that you can run it locally on your own computer using apps like To bebypassing all costs and privacy concerns altogether. That’s simply not possible with ChatGPT. If you’re an avid developer looking to integrate LLMs into your apps, Deepseek offers another benefit: significantly cheaper API access costs.
While there’s no flat-out winner yet, Deepseek is mostly free to try and can be run locally on your own computer, bypassing privacy concerns.
All that said, it’s early days for Deepseek specifically and for LLM models in general. In fact, the advent of a new model that can compete with the best and most funded in the business speaks volumes about the nascent state of the industry. Clever engineering can often bypass brute computational strength, and Deepseek points towards such an instance. Which one is better for you? I’d recommend trying out both and picking the one best suited for your needs. Me? I think I’ll be using Deepseek for a while longer till the next best thing comes out.
GIPHY App Key not set. Please check settings