I compared Opera Mini’s AI chatbot with ChatGPT and Gemini, and I’m impressed

Opera Mini is a mobile browser with a decadeslong legacy which predates the launch of even mobile platforms, including Android or iOS. Its popularity has since dwindled, especially as Google and Android offer more refined browsers, in Chrome and Safari, that are also preinstalled on phones. To keep up with this intense pressure, Opera — the eponymous browser company — has also taken some unwelcome steps, such as adding full-screen banner ads and a sensational news feed to its browser.

However, there is one area where Opera Mini is still relevant, and it is its data saving features, which used to be the selling point in the early days of smartphones when data was costly and limited. Though the issue like slow internet speeds or data caps are far less prevalent, we might still find ourselves stuck occasionally. And that’s why Opera Mini can still be useful.

More recently, Opera Mini was updated to include its AI chatbot, called Aria, which can answer queries or generate images. To see how really effective Aria is, I pit Opera Mini against ChatGPT and Google Gemini.

How Opera Mini’s AI is different

Foremost, Aria gets the convenience of being embedded in a web browser. If you already use Opera Mini, you don’t need a separate app. Aria primarily supports bimodal functionality, such that it supports text and images at the moment. Besides regular text queries and text-to-image generation, it also lets you attach images and answer questions based on that.

While these features are useful, they are also widely common among other chatbots. There is, however, one aspect that interests me. Opera says the chatbot, being part of the Opera Mini browser, also helps save data. In this article, I will be putting those claims to test, and checking if these data savings are truly worth it.

Unlike OpenAI, Google, Meta, and a whole wide range of providers have trained and developed their AI models ground-up, Opera says its chatbot is built on an AI engine that is powered by other models from Google and OpenAI. It doesn’t quite reveal the workings behind the said engine, so it would be interesting to see how the chatbot differs from actual offerings from these companies.

How I tested Opera’s AI against ChatGPT and Google Gemini

I split the test into four parts to test each chatbot’s acumen for commands of varying complexity. These tasks include testing Opera Mini’s Aria AI, Google Gemini, and ChatGPT for correctness, speed, and the data consumed for processing each request.

To measure the data consumed by each chatbot to process the various prompts, I pulled up Android’s app-wise data usage stats after running each prompt. While I reckon this isn’t the most ideal or scientific method of testing, the idea was to test Opera’s claims about data savings while using its AI, and if there’s any, get an estimate on the data consumed compared to the others. This is why, the data usage stats are rounded off in multiples of 0.1MB for simplification.

For consistency in results, I ran the same tests on free-to-use versions of Gemini (running 2.0 Flash model) and ChatGPT (GPT-4-turbo model).

Prompt 1: Basic text knowledge

For the first test, I chose a simple text prompt to examine each chatbot’s response time and data consumption with simple queries that may have already been fed to the AI model while training.

Here’s the first prompt I used:

What is the capital of Argentina?

Since the desired response isn’t contentious, all three chatbots respond correctly. Despite similarities in each one’s response, all three take different time to process the request and consume unequal amounts of data, as I have composed in the table below.

Both Gemini and ChatGPT were considerably faster at replying than Aria, which took almost four times as long as the other two. At the same time, Opera’s AI did not truly help save data in this test, especially in comparison to ChatGPT, which practically doesn’t consume any data to fetch the answer.

Despite a laggard start for Aria, let me take you through how it performed when challenged on other fronts.

Prompt 2: In-depth text processing with research

For the next text, I chose a more complex prompt that would require chatbots to analyze the problem in front of them before proceeding to generate the response. Besides testing them for quickness of response, the prompt also tests the chatbots’ abilities to draw information from various online sources.

The prompt I used for this purpose is as follows:

Create a detailed 5-day itinerary for a solo traveler visiting Kyoto, Japan for the first time. Include cultural experiences, food recommendations, and a rough daily schedule with travel tips.

The second prompt elicited more work than the first, so all the chatbots consistently took longer to process the requests. Here are the results compared in the similar fashion as the last one:

It is compelling to see ChatGPT pull up the information without using much data, as if it was doing so at the back of its head. While the cause isn’t immediately apparent, it feels ChatGPT may have already been trained for similar requests. It does, however, take the longest time to finish responding. Even though it starts generating the answer almost immediately, it continues to build it word-by-word. You may not bother with the delay, especially as only limited text is visible on a phone screen. By the time, you read and scroll, the latest sections would have populated, making the long duration to generate results less impactful.

Interestingly, the other two competitors take up much less time, and Opera’s Aria takes the least. As with ChatGPT, the responses generate word-by-word and load in the background, while you can begin scrolling. Aria also consumed much less data than Gemini, though there is a reason behind it — as well as ChatGPT’s lower consumption — which can be seen in the responses.

Among the three chatbots, ChatGPT had the most succinct responses. It took not more than a couple of sentences to list down the advised places to see while in the city of Kyoto. The instructions were simple and clear.

Gemini, on the other hand, offered an extensively detailed response, providing information on not just the tourist attractions and food recommendations — and what to expect there, but also suggested spots to buy souvenirs. The detailed itinerary felt useful, but the amount of information could also be overwhelming for some. Gemini’s answer might be more useful if I was planning the trip several months in advance, but not if I’m already in Kyoto or will arrive in a few days.

Lastly, Aria struck a balance between brevity and depth, breaking the days down in convenient chunks and organizing the information using solid headings, bullet points, and tables. Aria also mentioned its sources at the bottom, including not only blogs and webpages but also a YouTube video. I wish it also let me export the response in a Google Doc or PDF as Gemini and ChatGPT do. But overall, I was quite pleased with the answers from Aria.

Prompt 3: Image generation

For the third test, I inspected each chatbot’s abilities to generate images on command. Both Gemini and ChatGPT have already proven their mettle at the task and I did not expect them to slack. So, the test really was to scrutinize Aria.

This is the prompt I shared with the chatbots:

Generate a hyper-realistic digital painting of a futuristic city at dusk, viewed from a rooftop. The skyline should feature sleek glass skyscrapers with soft neon glows in purple and cyan. Flying vehicles are zipping between buildings. Add lush rooftop gardens with bioluminescent plants and a silhouetted figure watching the city below.

Notably, Opera also uses Google’s Imagen 3 text-to-image model as does Gemini. So, I expected both of them to produce similar results.

However, I was amused to learn Aria generated images much faster than Gemini, though both consumed similar amounts of data. The quality of the images — and the level to which both followed the prompt — were also identical. However, Gemini generated images that were twice in resolution as Aria, which explains the additional time.

ChatGPT was a laggard in this test, presumably owing to the heat its servers are having to endure due to the high amounts of image-generation requests. First with the Studio Ghibli effect and then with the action figure viral trends, multiple people have been turning to ChatGPT to create images with its improved image models. The phenomenon has even forced OpenAI to place restrictions for free users, which explains the long delay in image creation.

However, when it comes to skills, ChatGPT produced far more visually pleasing and refined results than both Gemini and Aria. You can see the results for yourself:

Image regeneration in different style

To challenge the chatbots’ abilities to recall previous prompts, I asked them to tweak the images generated above with another effect using the following prompt:

Recreate this image in an oil paint style. 

The time and data consumption for this recreation align with the previous test, as you can see below.

In fact, both Aria and Gemini consume less data, presumably because they do not have to recreate the entire image again. ChatGPT, on the other hand, takes in much more data — almost twice as the last time — and even more time. There is, however, a very good reason for that.

Both Aria and Gemini returned images without the desired modifications. Gemini simply recreated the image from a different perspective, while Aria degraded the image quality greatly and only adopted the requested oil painting effects for certain parts of the image. Even though ChatGPT consumed significantly more data and took way longer, it produced an image that truly looked like an oil painting without significantly modifying the elements in the image.

Prompt 4: Web browsing

Lastly, to test how well the chatbots fetch real-time information from the internet, I used a very specific prompt. Here’s what I asked:

Find the release date and standout features of the 2025 Fairphone, and briefly compare it to the previous model in terms of sustainability.

Contrary to the previous text-based requests where Opera Mini’s AI lagged the other, it was fast enough to fetch the information quicker and with minimal resources.

But what about accuracy? As it turns out, both Aria and ChatGPT (with its Search mode toggled on) gave a response based on the features of the older Fairphone 5, which launched in 2023, instead of the Fairphone 6, which is expected to launch in 2025. While ChatGPT completely misunderstood the question, Aria confidently said the product launched in January, which is actually when the company had shared their vision to launch a new phone this year.

Only Gemini was able to accurately point out that the 2025 model of Fairphone has only been teased but is yet to release. This suggests both ChatGPT and Aria have trouble understanding a slightly wordy and twisted question and tend to fetch results based on the first result they come across. Gemini, on the other hand, appears to be wary of the prompt’s wording and easily shows the right answer, which is likely a result of Google’s broader access to news articles, owing to its general dominance in web search.

Despite these flaws, I am inclined to recommend it to more people, especially those who don’t rely on AI for extensive or serious work.

Why I wouldn’t dismiss Opera Mini’s AI immediately

Opera Mini’s AI chatbot may lack any spectacular features or may not outshine Gemini, ChatGPT, or other chatbots by a large margin, but there is one reason to absolutely try it: lower data consumption. As I learned through a series of tests, the AI does seem to align with Opera Mini’s ethos of data saving. It probably wouldn’t make a significant difference if you use Wi-Fi for most of your adventures on the internet, or have a cellular data pack with unlimited — or almost unlimited — data balance.

But for instances when you are either low on data balance, facing issues with your internet’s speed, such as while using a hotel Wi-Fi, falling back to Opera Mini’s AI can be a reliable hack. Traveling internationally with limited data on your SIM is another scenario you may use Aria over the competitors.

I still hesitate in recommending you to use it solely, especially if you use AI for professional utilities. But if you want it to whip up a quick travel plan, or summarize a not-so-important web document, Aria can be trusted with it. And while AI may still be far from the technology that an average user heavily relies on, it’s good to have one that is embedded right into the web browser, as in Opera Mini’s case.

It still lacks compared to ChatGPT, especially since you can’t attach PDFs or other documents for it to summarize. But it does surprisingly well for a smallish model, especially in its ability to make these features available to the masses in developing or undeveloped countries.