How do I pick the right GenAI Model?

91% of people in my recent LinkedIn survey are using GenAI either occasionally or daily. A majority (54%) are using is daily, so workers are already embracing this tech to get their jobs done.

So what is a company to do? The first step is to facilitate cross-disciplinary whiteboarding with some GenAI experts at hand. Ideally, the ideas that come up there should then be packaged up as Proof of Concepts and shopped around with the large cloud providers (like AWS) to see if there is available funding to offset the implementation cost, along with calculating the total cost of ownership so you know what you are getting into long term.

If that sounds like a good idea, but like a lot of work, we happen to do this exact thing, at no cost to you (thank you uncle Bezos).

How many options are there?

When it comes to testing out and implementing GenAI, the open source platform Hugging Face is now tracking/hosting over 1.4M models—and is adding around 10,000 new models per week!

If this feels overwhelming, it's because it is. Also, because of the buzz around AI right now a LOT of what's being put out there is absolute trash. Until you experiment with something directly it's tough to tell the jewels apart from the junk, but there are certainly what we would call major players.

GenAI Model Tracking

In order to help me wrap my head around the landscape, I created a spreadsheet, starting with the models that AWS runs on Bedrock, their GenAI platform (think of Bedrock as the hardware to the model software and you're mostly there) and then adding the the big competitors: Google & Microsoft. AWS is pretty quick at adding in the cream of the crop (like DeepSeek and Claude 3.7) and I do my best to keep the spreadsheet up-to-date:

AWS Bedrock Advantage

When you compare AWS's GenAI approach to their big competition as few things stand out:

1. AWS  gives you choice. Unlike Microsoft & Google, on Bedrock you can run over 60 different models and counting. This includes Llama by Meta (Facebook)
2. AWS is cost effective. AWS GenAi models run at a fraction of the price of some of the others. To quote a user on Reddit regarding the 4.5 OpenAI release: "This is the the kind of pricing you'd offer for something you didn't really want people using."
3. AWS responds quickly. When something new hits the scene (like DeepSeek) AWS is on top of their game and has integrated the new models in as quickly .
   

I've said this before, we are still at the beginning, but every industry will be transformed, and it remains to be seen who will come out on top. There's no question that our relationship with AI will continue to evolve, but even with the tools we have today, there is plenty to be done. We'd be happy to help.