What is our primary use case?
I am aware of Google Gemini translation tools, but I haven't used them. They don't impact my work as all my work is done in English, so I don't need any kind of translation.
I've been using the API model of Google Gemini to do integrations. I use the tool to make decisions on the next step through the API to Google Gemini, which helps make my software smarter. I can make decisions based on what kind of information I need to extract and parse. Based on Google Gemini, I make my software easier to integrate and easier to make decisions, because I don't have to think about all the details. I just focus on what is important for me.
What is most valuable?
There is another benefit in that Google Gemini is highly competitive from a cost perspective. The cost per token is very competitive. It's not the best - the Chinese models such as DeepSeek are better, but I can't use them. From a cost perspective, between European and United States options, Google Gemini is very competitive.
It's not only its cost, it's also very performing. The time it takes to process the prompts and context information is very fast. It's one of the fastest models in the market. That's another advantage. There are other models which are also very good in quality but not as fast, such as Anthropic Claude. Google Gemini currently has a very good balance.
The AI capabilities of Google Gemini are a multi-modal LLM which allows me to pass documents, images, and texts in the same prompt. Being multi-modal makes my life easier. I don't have to translate information before I can work with it. Google provides free tools to test scenarios, which speeds up development and reduces costs. It's a very fast and accurate model with quality responses.
What needs improvement?
There's a new release every two months and it's very fast. I would say that they should keep increasing the context size of information that we can provide. Going from 1 million to 10 million would be fantastic. The context size and improving creativity without hallucinating is very difficult to do, but I expect them to achieve that. I expect them to be more creative but still control the amount of hallucination.
For how long have I used the solution?
The tool just works as expected.
What was my experience with deployment of the solution?
The tool just works as expected.
What do I think about the stability of the solution?
The tool just works as expected.
What do I think about the scalability of the solution?
The biggest blocker for scaling is the context size. If you want to grow the amount of information that you want to insert into the model before you provide an answer, you have to use different techniques. The context window is always a limitation. There are techniques which help you to address those limitations, retrieve augmented generation, RAG, and Kada. That's where the problems are too complex, and you have to find ways to handle the scale of the problem. There are scalability issues, and they are all related to what you can do within that context window of a million tokens.
Regarding ease of scaling, I would rate it a six.
How are customer service and support?
So far, it has been perfect, deserving a rating of ten.
How would you rate customer service and support?
Which solution did I use previously and why did I switch?
I have used OpenAI, specifically the GPT-3 Mini and other models from OpenAI, including ChatGPT. I have also used the Anthropic model called Sonnet.
How was the initial setup?
It is very easy to set up as it's a well-thought tool for developers.
What about the implementation team?
You create your own application and your application can use Google Gemini by calling directly the Google Cloud. You don't have to worry about development, and concerning the Gemini tool, you don't have to worry about having multiple environments for development, quality and production. You just use a single one.
What was our ROI?
There is a free layer which you can use as a developer, using the Google AI Studio tool. For some of the models it's actually free. It doesn't cost anything, but once you get to production scenarios in which you have to use the API, you have to pay. If you're using the APIs to make calls for automatic scenarios of data analysis, then you have to pay. It's very competitive. In a scale of one to ten in the market, I would rate it probably a seven or eight.
What's my experience with pricing, setup cost, and licensing?
If ten is expensive and one is free, I would rate it a three or four.
Which other solutions did I evaluate?
Google Gemini has the best combination of scalability, costs, performance, and accuracy. These things evolve very fast, but currently, on these four vectors, I believe it's very competitive and the combination of these four is better than any other that I'm aware of.
What other advice do I have?
To hallucinate is when the model lies to you. Sometimes the responses are lies because it starts to imagine things which are not real. That's a hallucination. It's similar to being on drugs. I would rate Google Gemini as an eight.
Which deployment model are you using for this solution?
Public Cloud
If public cloud, private cloud, or hybrid cloud, which cloud provider do you use?
Google
Disclosure: I am a real user, and this review is based on my own experience and opinions.
