An In-Depth Practical Gemini 2.5 Test
10 min read
Google's AI covers a huge amount of ground these days, tackling everything from complex analysis to creative tasks. Before you point out imperfections, it's crucial to make sure you've instructed it exactly what you need and outlined the task precisely how you want it done. Like any powerful tool, clear direction gets the best results, and often a bit of refinement is part of the process to get things just right.
Google is now leading the AI race in terms of having the most powerful and intelligent tools. The Gemini Flash (available for all) and Gemini Pro (available for subscribers) models are indeed amazing. From deep analysis and complex tasks, to helping you write that song for your loved one, Google will help you all the way. So far, I haven't run out of tokens, it feels like it just goes on until the task is done or I've achieved what I'm aiming for.
Google's AI models can analyze text, videos, audio, and images, and their built-in research capabilities can even do deep research on the Internet for you. This means gathering information from the internet and turn it into valuable fact-checking, competitive analysis, or identifying potential resources. Basically, there are almost no limits to what you can research. I really like that, because analysis gets you far when you study the fundamentals before putting a plan into motion.
I live in Denmark, so I thought it would be great to catch up on how well we're doing on tourism here. I used Google's AI research features to help collect the data for this article. Note that Google doesn't collect all data on the Internet to do its research. It uses the sources most likely to have this kind of information, however, you can instruct it to include certain elements which you are sure it might not collect.
Model Overview
Benchmark Comparison
Before we do our report analysis, let's take a look at the benchmark comparison between the two models. The table below shows the key features of both models, including their input context window, maximum output tokens, and more.
Feature
GPT-4o
Gemini 2.5 Flash
Input Context Window
The number of tokens supported by the input context window.
128K tokens
1M tokens
Maximum Output Tokens
The number of tokens that can be generated by the model in a single request.
16.4K tokens
65K tokens
Open Source
Whether the model's code is available for public use.
No
No
Release Date
When the model was first released.
August 6, 2024
9 months ago
April 17, 2025
1 month ago
Knowledge Cut-off Date
When the model's knowledge was last updated.
October 2023
January 2025
API Providers
The providers that offer this model. (This is not an exhaustive list.)
OpenAI, Azure OpenAI Service
Google AI Studio, Vertex AI, Gemini app
Supported Modalities
The types of inputs the model can process.
Understanding the AI Model Comparison
The benchmark overview compares two leading AI models. OpenAI's GPT-4o and Google's Gemini 2.5 Flash, across several key technical features. Understanding these features helps clarify each model's strengths and potential use cases. Here's a breakdown:
Input Context Window
This is like the AI's short-term memory. The maximum amount of information (text, code, previous conversation) it can consider at once when generating a response. Larger numbers mean it can handle longer documents or conversations without forgetting earlier parts.
Maximum Output Tokens
This defines the longest possible response the AI can generate in a single request. A higher limit allows for longer articles, more extensive code generation, or more detailed explanations in one go.
Open Source
Indicates if the model's source code and potentially its trained weights are publicly available for inspection, modification, and redistribution by the community.
Release Date
The date when this specific version of the AI model was first officially announced or made available to users or developers.
Knowledge Cut-off Date
Represents the point in time when the model's training data ended. It generally won't have knowledge of events, developments, or information published after this date.
API Providers
These are the platforms or services where developers can access the model programmatically to build their own applications. (List may not be exhaustive).
Supported Modalities
The different types or formats of information the AI model can understand as input or generate as output.
In Summary: This comparison highlights key differences. Gemini 2.5 Flash boasts a significantly larger context window (allowing it to process much more information at once) and higher maximum output length compared to GPT-4o. However, GPT-4o had an earlier release and a different knowledge cut-off date. Both models are closed-source but available through major cloud/API providers and support multiple input/output types (text, image, audio, video).
Putting Google's AI to the Test
This entire analysis following below, from researching Denmark's 2024 tourism statistics to generating the charts and layout components you see on this page, was built in collaboration with Google's Gemini 2.5 models. Putting these tools to a practical test like this really highlights their current strengths and limitations.
So, how did it do? Honestly, it was impressive but not flawless. I'd estimate that Gemini handled about 60% of the heavy lifting, significantly speeding up the process.
Despite the need for refinement, the ability of Gemini 2.5 to understand the design constraints and generate code that largely fit the existing structure was a massive time-saver.
While Gemini 2.5 already makes tasks like this significantly easier, the potential leap to a "3.0" generation could further blur the lines between AI assistance and genuine AI collaboration on complex web development and data analysis projects. The speed of progress remains astounding.
Interactive Map of Landmarks in Denmark
Use mouse scroll or pinch-to-zoom to navigate the map.
Key Figures for Denmark Tourism 2024
65.2 M
Total Overnight Stays
+2.3% vs 2023
34.2 M
International Stays
+5.2% vs 2023
30.9 M
Domestic Stays
-0.8% vs 2023
21+ M
German Stays
+4.4% vs 2023
Top International Markets by Overnight Stays (2024)
Approximate share based on reported figures.
Tivoli Gardens
Quarterly Visitor Distribution (2024)
Total Visitors: 4.25 Million (+5% vs 2023). Note the strong Q4 performance.
Tivoli Gardens Visitor Mix (2024 - Approx.)
Based on 4.25M total visitors, with >1.4M international. Domestic calculated.
Landmark Data Availability (2024)
Landmark
Annual
Quarterly
Monthly
Tivoli Gardens
Yes
Yes
No
Legoland Billund
Policy
Policy
Policy
ARoS Aarhus
No (2023 Avail.)
No
No
Kronborg Castle
No (2023 Avail.)
No
No
Rosenborg Castle
No
No
No
Nyhavn / Little Mermaid / Grenen
N/A
N/A
N/A
N/A: Not Applicable (Public/Natural Site). Policy: Data not publicly disclosed. No: Data not found in sources for 2024.
Overnight Stays
Domestic vs. International (2024)
Based on 35.86 Million total stays reported Jan-Oct 2024.
Tourism Growth
Copenhagen vs. National Average (2024 vs 2023)
Comparing % growth in overnight stays.
Estimated International Arrival Modes (Conceptual)
Conceptual visualization based on report (Car inferred, Air proxy, Ferry peaks). Rail data unavailable.
Report Summary
Denmark's tourism sector achieved a record 39.9 million overnight stays in 2024, a 2% increase from 2023, primarily driven by strong international visitor growth (+9% arrivals to accommodations). While domestic stays formed the majority (60.3% in Jan-Oct), international tourism, led by Germany, the Netherlands, and Norway, was the key growth engine.
Copenhagen significantly outperformed the national average, with overnight stays rising 7% to 11.1 million, 7 million of which were international. This highlights the capital's dominance, further evidenced by Tivoli Gardens being the most visited landmark with verifiable 2024 figures, attracting 4.25 million visitors (+5% YoY).
While showing overall growth and resilience surpassing pre-pandemic levels in some metrics, the analysis faced data limitations. Comprehensive monthly breakdowns for 2024, specific visitor counts for many landmarks (due to policy or public nature), and detailed transport mode statistics were unavailable in the provided sources. However, available data confirms a strong summer peak, significant air and peak ferry traffic, and the implied importance of car travel via the German market.
The overall outlook remains positive, supported by Denmark's strong international reputation and global tourism growth forecasts, although future performance depends on economic and geopolitical factors.
Conclusion
I am amazed to see what AI is capable of doing, but also a bit scared. I've worked with quite a few AI projects now through various communities, and the speed of everything is certainly something that shouldn't be overlooked. We really need to see strong European collaboration to compete on this level, because there's no doubt that the big tech companies in the US, and the growing AI companies in China, are already setting their mark, and they are not slowing down.
Google's advantage in training their models using their own systems and hardware makes them very competitive and likely to take the lead entirely. However, Microsoft with its Azure architecture, Meta, Amazon with its AWS architecture can still keep up, but for how long? For AI tools to become more intelligent, they need more hardware for more extensive training. Also, on the consumer level, they need the server power to support users when they want help solving the next problem or writing the next song. Running these Large Language Models requires a lot of power. Optimizing LLMs and figuring out different architectures for them to work on is definitely a hot topic.
Let's not forget that the Ford Model T was built as a conventional car to get you from A to B, and it took a few years before we saw the first race car. But the speed of AI far outperforms the speed of an engineer from that time, so we don't have the luxury of waiting.
Speaking as a European, less talk, more action, more visibility.
Discussion
Join the conversation! Please log in or sign up to comment.