Hi all,
A quick blog post to announce a pretty big step forward for 4AI in version 5.0. Since its appearance nearly two years ago, 4AI has worked by using the OpenAI AI service. OpenAI are the makers of ChatGPT and at the time, they were not only the obvious choice as a provider for an AI API, but essentially the only one.
Google Gemini models in 4AI
I've been continuously monitoring and testing all the various API providers which have appeared over the last 2 years: Claude, Mistral and of course DeepSeek just recently.
None of them appeared valid choices to be part of 4AI, either for reasons of price, availability, reliability or just speed.
This changed on February 5, when Google released Gemini version 2.0. Gemini is fast, much faster than OpenAI, it works very well and is essentially free for use in 4AI, as Google's free tier allows up to 1500 requests per day free of charge.
What's the catch?
Looks like there's always a catch! For Google Gemini, there are two:
Free-tier data can be used for training
The biggest catch: if you use the free tier (i.e. no credit card, just use up to 1500 requests per day), then your data (questions you ask and responses you get) can be "Used to improve our products". Not sure what that means in practice but you have to be aware of this and decide if you want to use the free of charge version, or prefer paying.
Note that if you opt for the paying tier, Google Gemini pricing is still around 10 times lower than OpenAI's.
No images. Yet.
At this point, 4AI does not integrate image generation for Google Gemini. When development started, image generation was not available yet from Google. It only became available recently, and will be added to 4AI in the coming weeks.
Note that image generation with Google Gemini is not free. Current cost is slightly lower than the OpenAI equivalent, with excellent control on the aspect ratio of generated images.
What's good?
When Google Gemini 2.0 was released, my testing showed that it brings a notable improvement over current OpenAI models for all common operations. If anything, the speed is much better and all text results are at least on par.
And of course, there's the price, which allows anyone to use a high-end AI model at no cost - as long as you can live with the privacy caveat, or very little cost if privacy is essential.
I look forward to adding Gemini image generation to 4AI, and of course support for more competing models in the near future.
What else?
4AI version 5 also adds support for the latest OpenAI models: o1, o1-mini and o3-mini.
o3-mini is now the default if you select OpenAI as a provider for a given function, as it's the best combination of price, "intelligence" and speed. Of course, you can always select another model based on the work at hand.
Cheers,
Yannick