Google’s Gemini just dropped 3 cutting edge new models. This morning, I tried them all.

Are these new models really better? Let’s find out…

2.0 Flash Thinking Experimental

This new model is designed for multi-step reasoning.

I’ve been trying to learn about missile defense recently. So, I had the model evaluate ways to stop incoming missiles. This should be a good test of multi-step reasoning.

Here’s the prompt I used:

“I’m interested in new technologies that could help the United States stop incoming missiles, in particular ICBM and hypersonic missiles. What possible new technologies would be best suited to this task? Evaluate each possible new technology for viability.”

Gemini gives us a reasoning trace, explaining how it will attack the problem. The trace is similar to how DeepSeek R1 answers questions.

Gemini’s answer is fascinating, covering everything from directed energy weapons to cyber warfare.

It’s a great overview, but I would’ve loved to see some links to more information. That helps me evaluate how accurate Gemini’s statements are. You can “Double Check Response,” but I’d rather just see the citations up front.

Overall, I’m giving this model a B+.

2.0 Flash Thinking Experimental with apps

Google’s second new model is designed to reason using information in YouTube, Google Maps, and search.

This power is what makes Google so dangerous. They have all the tools and can use them with AI to provide a better user experience.

I asked Gemini to help me find a hotel for a trip to Tokyo that I’m planning. This should make good use of the Google Maps integration. Here’s the prompt I used:

“If I’m planning a fall trip to Tokyo, which locations of the hotel Toyoko Inn have the best price? in particular, I’m interested in locations near a major train station, if possible.

Tell me the best locations and give me prices for each one.”

Gemini’s answer wasn’t very helpful. It provided a nice list of hotels, but no prices.

It looks like the prices for fall aren’t available yet. But a better response would’ve included the typical rates at that time of year.

I’ll give this response a C.

2.0 Pro Experimental

Lastly, we have the Granddaddy of Gemini. This model is for complex tasks.

I had the model attack a difficult problem: finding startups that could increase American fertility. I gave it a complex prompt, searching for very specific types of companies:

“I’m an angel investor interested in finding startups that can help increase fertility in the United States. Some possible approaches these startups could take: 1) Extend women’s fertility with drugs or medical procedures. 2) Make IVF more affordable and effective. 3) Make it easier to live and work in suburban and rural areas, as opposed to cities. (Urbanization correlates negatively with fertility.) 4) Promote religion (religiosity correlates with fertility). Please find me startups that take any of those approaches. Also consider startups that take other approaches to increasing fertility that I may not have thought of. When you find a list of startups, narrow it down to those that have raised $750,000 or less. You can find data on fundraising on Crunchbase, Pitchbook, etc. Give me the final list of startups in a table. Include the startup’s name, what the startup does, how much money it has raised, when it was founded, and where the startup is based.”

Here’s what Gemini found:

Gemini does a nice job of searching for startups and giving me the info in a readable format. I checked the information and it’s mostly correct, although sometimes Gemini got the funding amount or founded date wrong.

I’m looking deeper into these companies to see if I can find an investment here…

I’ll give this response an A-.

Wrap-Up

Gemini’s new models are great at analyzing mountains of data and pulling out useful info. They’re also awesome at tutoring, as we saw with the missile defense example.

But the integration with other Google services and the internet as a whole could use some work. Simple things like pulling hotel prices or startup funding amounts sometimes don’t work.

Overall, I’m giving Gemini’s new models a B+. They’ve got some great features but still need to work on accuracy.

I still think 1.5 Pro with Deep Research is the most impressive Gemini model. It’s on par with OpenAI’s new Deep Research at 1/10th the cost with no rate limits.

This is the best time to be in tech in the last 30 years. Every day, companies are releasing incredible new tools.

What a time to be alive!

More on tech:

DeepSeek vs. Gemini Deep Research: Which Model Is King?

ChatGPT Pro vs. Gemini Advanced vs. Grok vs. Claude

Request for Startups – February 2025

Save Money on Stuff I Use:

Fundrise

This platform lets me diversify my real estate investments so I’m not too exposed to any one market. I’ve invested since 2018 with great returns.

More on Fundrise in this post.

If you decide to invest in Fundrise, you can use this link to get $100 in free bonus shares!

Misfits Market

I’ve used Misfits for years, and it never disappoints! Every fruit and vegetable is organic, super fresh, and packed with flavor!

I wrote a detailed review of Misfits here.

Use this link to sign up and you’ll save $15 on your first order.