I Tested DeepSeek’s R1 and V3 Coding Skills – and we’re not All Doomed (Yet).

DeepSeek took off into the world’s awareness this previous weekend. It stands apart for three powerful factors:

1. It’s an AI from China, instead of the US

2. It’s open source.

3. It uses vastly less facilities than the big AI tools we have actually been looking at.

Also: Apple researchers expose the secret sauce behind DeepSeek AI

Given the US government’s concerns over TikTok and possible Chinese government participation because code, a new AI emerging from China is bound to produce attention. ZDNET’s Radhika Rajkumar did a deep dive into those problems in her post Why China’s DeepSeek might rupture our AI bubble.

In this article, we’re preventing politics. Instead, I’m putting both DeepSeek V3 and DeekSeek R1 through the same set of AI coding tests I have actually tossed at 10 other large language models. According to DeepSeek itself:

Choose V3 for tasks requiring depth and precision (e.g., resolving innovative math issues, producing intricate code).

Choose R1 for latency-sensitive, high-volume applications (e.g., consumer assistance automation, basic text processing).

You can select between R1 and V3 by clicking the little button in the chat user interface. If the button is blue, you’re utilizing R1.

The brief response is this: outstanding, however plainly not best. Let’s dig in.

Test 1: Writing a WordPress plugin

This test was in fact my very first test of ChatGPT’s programs expertise, way back in the day. My spouse needed a plugin for WordPress that would help her run a participation device for her online group.

Also: The very best AI for coding in 2025 (and what not to utilize)

Her requirements were fairly simple. It needed to take in a list of names, one name per line. It then had to arrange the names, and if there were duplicate names, different them so they weren’t listed side-by-side.

I didn’t truly have time to code it for her, so I chose to give the AI the challenge on an impulse. To my substantial surprise, it worked.

Ever since, it’s been my first test for AIs when evaluating their programming skills. It needs the AI to understand how to set up code for the WordPress structure and follow prompts clearly sufficient to create both the interface and program logic.

Only about half of the AIs I’ve tested can totally pass this test. Now, however, we can include one more to the winner’s circle.

DeepSeek V3 produced both the user interface and program logic exactly as defined. As for DeepSeek R1, well that’s a fascinating case. The “thinking” element of R1 triggered the AI to spit out 4502 words of analysis before sharing the code.

The UI looked various, with much larger input locations. However, both the UI and reasoning worked, so R1 also passes this test.

Up until now, DeepSeek V3 and R1 both passed one of 4 tests.

Test 2: Rewriting a string function

A user grumbled that he was not able to get in dollars and cents into a contribution entry field. As written, my code just enabled dollars. So, the test involves giving the AI the routine that I composed and asking it to reword it to permit both dollars and cents

Also: My preferred ChatGPT feature just got way more effective

Usually, this results in the AI creating some routine expression validation code. DeepSeek did create code that works, although there is space for enhancement. The code that DeepSeek V2 composed was unnecessarily long and repetitive while the thinking before creating the code in R1 was also extremely long.

My biggest concern is that both designs of the DeepSeek validation ensures validation up to 2 decimal locations, however if a huge number is gotten in (like 0.30000000000000004), making use of parseFloat doesn’t have explicit rounding understanding. The R1 model also utilized JavaScript’s Number conversion without inspecting for edge case inputs. If bad information comes back from an earlier part of the regular expression or a non-string makes it into that conversion, the code would crash.

It’s odd, because R1 did provide an extremely good list of tests to validate versus:

So here, we have a split choice. I’m providing the indicate DeepSeek V3 due to the fact that neither of these issues its code produced would cause the program to break when run by a user and would create the expected results. On the other hand, I have to provide a fail to R1 due to the fact that if something that’s not a string somehow gets into the Number function, a crash will take place.

Which offers DeepSeek V3 2 triumphes of 4, however DeepSeek R1 just one triumph of 4 so far.

Test 3: Finding an irritating bug

This is a test created when I had a really irritating bug that I had problem locating. Once again, I decided to see if ChatGPT could manage it, which it did.

The difficulty is that the answer isn’t apparent. Actually, the difficulty is that there is an apparent response, based on the mistake message. But the apparent response is the wrong response. This not only captured me, but it routinely catches some of the AIs.

Also: Are ChatGPT Plus or Pro worth it? Here’s how they compare to the totally free variation

Solving this bug needs understanding how specific API calls within WordPress work, being able to see beyond the mistake message to the code itself, and then understanding where to discover the bug.

Both DeepSeek V3 and R1 passed this one with nearly identical responses, bringing us to 3 out of 4 wins for V3 and 2 out of four wins for R1. That already puts DeepSeek ahead of Gemini, Copilot, Claude, and Meta.

Will DeepSeek score a crowning achievement for V3? Let’s discover.

Test 4: Writing a script

And another one bites the dust. This is a tough test since it needs the AI to understand the interplay between 3 environments: AppleScript, the Chrome things model, and a Mac scripting tool called Keyboard Maestro.

I would have called this an unjust test because Keyboard Maestro is not a traditional shows tool. But ChatGPT dealt with the test easily, comprehending exactly what part of the problem is managed by each tool.

Also: How ChatGPT scanned 170k lines of code in seconds, saving me hours of work

Unfortunately, neither DeepSeek V3 or R1 had this level of understanding. Neither model understood that it required to split the job between guidelines to Keyboard Maestro and Chrome. It likewise had relatively weak understanding of AppleScript, composing customized routines for AppleScript that are native to the language.

Weirdly, the R1 model stopped working too due to the fact that it made a lot of inaccurate assumptions. It presumed that a front window constantly exists, which is definitely not the case. It likewise made the assumption that the presently front running program would constantly be Chrome, rather than clearly checking to see if Chrome was running.

This leaves DeepSeek V3 with three right tests and one stop working and DeepSeek R1 with 2 proper tests and 2 stops working.

Final ideas

I discovered that DeepSeek’s insistence on utilizing a public cloud email address like gmail.com (instead of my regular email address with my business domain) was frustrating. It likewise had a number of responsiveness fails that made doing these tests take longer than I would have liked.

Also: How to utilize ChatGPT to compose code: What it succeeds and what it doesn’t

I wasn’t sure I ‘d be able to write this post because, for most of the day, I got this mistake when attempting to register:

DeepSeek’s online services have actually recently faced large-scale harmful attacks. To ensure continued service, registration is momentarily limited to +86 phone numbers. Existing users can log in as normal. Thanks for your understanding and assistance.

Then, I got in and was able to run the tests.

DeepSeek seems to be overly chatty in regards to the code it produces. The AppleScript code in Test 4 was both wrong and excessively long. The routine expression code in Test 2 was correct in V3, but it could have been written in a way that made it far more maintainable. It stopped working in R1.

Also: If ChatGPT produces AI-generated code for your app, who does it actually belong to?

I’m certainly impressed that DeepSeek V3 vanquished Gemini, Copilot, and Meta. But it seems at the old GPT-3.5 level, which means there’s certainly room for improvement. I was dissatisfied with the results for the R1 design. Given the choice, I ‘d still select ChatGPT as my programs code helper.

That said, for a brand-new tool running on much lower infrastructure than the other tools, this could be an AI to enjoy.

What do you think? Have you tried DeepSeek? Are you using any AIs for shows support? Let us understand in the remarks below.

You can follow my everyday task updates on social networks. Make certain to register for my weekly upgrade newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/ DavidGewirtz, on Instagram at Instagram.com/ DavidGewirtz, on Bluesky at @DavidGewirtz. com, and on YouTube at YouTube.com/ DavidGewirtzTV.