Gemini 4 Argon vs. GPT-6 wasn’t the showdown I anticipated

Google lately launched Gemini 4 Argon, and it appears to be like extremely succesful. It’s designed to deal with all the pieces from complicated coding duties and cybersecurity protection to robotics and long-term reasoning issues.

Now, I already pay for AI subscriptions and use these fashions just about on daily basis, each for work and private duties. So, I used to be curious to see what Google’s newest mannequin might really do. More importantly, I needed to see the way it holds up in opposition to OpenAI’s GPT-6, which has turn into my go-to AI mannequin for many issues.

I made two AIs put a cat on wheels

One made it fairly, the opposite made it playable

I began with one thing enjoyable — I requested each Gemini 4 Argon and GPT-6 Sol to construct a sport I might open and play immediately in my browser, with out putting in something or coping with a number of recordsdata. And to maintain issues honest, I gave each fashions the very same immediate:

Make a small, enjoyable sport in a single HTML file that I can open in my browser. A cat on a skateboard rolls left and proper to catch falling stars and dodge rain clouds. Add a rating, progressively enhance the velocity, present a game-over display screen with a restart button, and make it work with each arrow keys and contact on a cellphone.

I wasn’t anticipating something notably fancy, however I needed to see how effectively each fashions might flip a easy thought into one thing I’d really take pleasure in enjoying. Gemini 4 Argon completed the job in slightly below three minutes, whereas GPT-6 Sol took a little bit longer, clocking in at precisely three minutes and 37 seconds.

One factor I instantly preferred about Gemini was the way it dealt with the whole course of. I might see a reside preview of the sport because it was being constructed, together with the code operating within the background. It made the entire expertise really feel far more interactive, whereas ChatGPT did not provide the identical comfort — it talked about the steps, sure, however nothing extra.

However, issues obtained attention-grabbing as soon as I really began enjoying the sport. Gemini had gone with an enthralling retro-style design, which I assumed suited the entire cat-on-a-skateboard thought completely. Unfortunately, the controls `weren’t almost nearly as good because the visuals. It gave me two directional arrows to maneuver the cat round, and whereas they labored moderately effectively at first, controlling the cat grew to become more and more irritating as the sport picked up velocity. I struggled extra with the controls than with catching stars or dodging clouds.

ChatGPT, alternatively, took a a lot less complicated method. Its sport did not look almost as thrilling as Gemini’s, and the general design was fairly minimal. But as soon as I began enjoying, I shortly realized that appears weren’t all the pieces.

Instead of counting on two directional buttons, GPT-6 Sol gave me a slider to maneuver the cat left and proper. It may sound like a small distinction, but it surely made the sport a lot simpler to regulate, particularly as soon as all the pieces began shifting sooner. So, whereas Gemini 4 Argon was faster and made the better-looking sport, GPT-6 Sol received me over with how effectively it really performed.

I additionally let it plan my electrical escape

But Gemini saved me on monitor

For my second take a look at, I made issues a little bit extra difficult. This time, I needed each AIs to assist me plan a 600 km electrical street journey, determining the place I ought to cease to cost and the way a lot time I’d want at every station.

It wasn’t an extremely tough drawback, however in comparison with asking an AI to construct a easy sport, I assumed this may be a great way to check how effectively each fashions might deal with numbers and clarify their reasoning. This was my immediate:

I’m taking a 600 km freeway journey in an electrical automobile. The battery begins at 100%, and each 1% will get me 2.5 km. I by no means wish to drop beneath 10%, together with on arrival. Chargers are at 120km, 200, 310, 400, and 480km. Chargers add 1% per minute as much as 80%, then sluggish to 1% each 4 minutes. The charger at 310km is older and solely provides 1% each 2 minutes, as much as 80%. What’s the least whole time I can spend charging, and the place ought to I cease? Explain in plain language.

Now, I requested each to clarify all the pieces in plain language for a cause. I’m not notably good with numbers, and when a proof begins throwing too many calculations at me, I are inclined to lose monitor.

Interestingly, each fashions completed the duty in roughly 32 seconds and gave me the identical right reply. They additionally organized their responses into tables, which made it simpler to see the place I wanted to cease and the way lengthy I’d be charging. So, so far as fixing the precise drawback was involved, there wasn’t a lot separating the 2.

But as I learn by their explanations, I seen a distinction. Gemini 4 Argon appeared to grasp precisely what I meant by “plain language.” Instead of overwhelming me with calculations, it broke all the pieces down into smaller, easier-to-follow sections, used two units of pointers to clarify the reasoning, and included a desk that tied all the pieces collectively. ChatGPT did an excellent job, too. But its clarification wasn’t as detailed.

And that is what made the distinction for me. So, for this spherical, I’d fortunately give Gemini 4 Argon the win.

I intentionally planted seven traps for his or her brains

Both walked away unscathed

For my closing take a look at, I needed to push each fashions a little bit more durable. So, I moved on to one thing far more technical: a safety audit of a Flask software. Now, I’m not a cybersecurity professional, however I needed to see how effectively each AIs might establish safety vulnerabilities, clarify why they had been harmful, and repair them with out introducing new issues.

I gave each fashions the identical piece of code, which had seven intentionally planted safety flaws, together with this immediate:

Review this Flask app as a safety auditor. List each vulnerability, rank them by severity, clarify in a single line how every could possibly be exploited, then give a hard and fast model of the complete file. Don’t pad with generic recommendation.

Unlike the earlier take a look at, this wasn’t nearly getting a solution or making one thing enjoyable. I needed them to rigorously study the code, catch the vulnerabilities I’d included, and provides me a corrected model that really addressed these issues. ChatGPT completed the duty in 4 minutes and 18 seconds, whereas Gemini took almost 5 minutes. That’s lower than a minute, which I would not contemplate notably vital for a job this difficult.

As I’d come to anticipate by this level, Gemini 4 Argon went all out with its clarification. It broke down the vulnerabilities in appreciable element, defined how every one might probably be exploited, and walked me by the fixes. GPT-6 Sol wasn’t far behind, both. Its response was additionally detailed, well-organized, and simple sufficient to observe.

And that is why I could not actually choose a winner right here. Gemini was a little bit extra thorough in its explanations, whereas ChatGPT completed barely sooner, however neither had a significant benefit on the precise job. Both caught all seven planted flaws and delivered what I’d requested for, so I’m calling this spherical a tie.

Three assessments later, my loyalty to AI is difficult

After spending time with each fashions, I’ve realized selecting the higher AI is not almost as simple as I assumed. I’ve been using ChatGPT for quite some time now, and truthfully, I went into these assessments anticipating it to come back out forward. But Gemini surprised me more than once, particularly with the way it introduced info and made difficult issues simpler to grasp.

That mentioned, I’m positive there are many conditions the place the outcomes might go both approach. For now, I’m not planning to change away from ChatGPT, largely as a result of I’m snug with it, and it already matches properly into my every day workflow. But Gemini 4 Argon has undoubtedly given me a cause to maintain coming again and experimenting.