I Tested Gemini 3.8 Flash, Google’s Latest AI Update, and It (Mostly) Lives up to Expectations.
Google has released two new AI models : Gemini 3.8 Flash (an “intelligent workhorse” for everyday suggestions) and Gemini 3.8 Flash Cyber (available to “trusted professionals” for cybersecurity protection). Although Gemini 3.7 was released just a few weeks ago, the new model brings significant improvements.
According to Google, Gemini 3.8 Flash delivers better reasoning and programming capabilities than ever before, with “significant” improvements in “software development, agent-based tasks, and critical multi-step reasoning in specialized domains”—making this update more geared toward advanced AI than basic tasks. Gemini 3.8 Flash is available now in the Gemini app if you’re on the AI Pro ($19.99/month) or AI Ultra ($99.99/month) plan. There’s no update yet for free or $7.99/month AI Plus users, though these plans will gradually become available on lower plans as updates are released.
Gemini Flash is a new and improved version.
It’s understandable that you might be a little confused by the plethora of Gemini AI models Google has released in recent months. For some time now, these models have been categorized as Flash-Lite, Flash, and Pro, and the higher the level of AI, the slower it operates, the more sophisticated its reasoning and complex query capabilities, and the more expensive it is. (Not to mention that these models have different levels of intelligence .)
In the Pro category, we’re still using Gemini 3.1 Pro, released in February 2026. Although a sequel was mentioned at Google I/O in May , it hasn’t been released yet. The Flash channel is developing much faster, and only a few months ago, in December, the first preview of Gemini Flash 3.0 was presented.
Now we have Gemini 3.8 Flash. According to Google’s tests (which should always be taken with a grain of salt), it offers improvements over its predecessor in solving “long-term software development” problems (i.e., full-fledged projects, not small tools), information processing (data synthesis and analysis), complex legal processes, financial analysis, and agency work. According to Google, Gemini 3.8 Flash is particularly good at solving complex engineering problems and is improved at multi-step reasoning in science, mathematics, humanities, and professional fields. It will analyze answers more deeply, perform more accuracy checks, and better combine different skills in a single solution.
There are also security measures in place: the AI has mechanisms to protect against misuse of questions related to chemical, biological, radiological, nuclear weapons, and cybercrime, and is better than ever at resisting hint-based attacks (where hints are designed to trick the AI into doing something malicious). At first glance, everything looks impressive, as do the demos Google is showing here: one of them is a fully working version of Google Maps in the MS-DOS interface , which will be especially appealing to those (like me) who started using computers back in the days of the command line.
Testing Gemini 3.8 firmware in comparison with Gemini 3.6 firmware.
While I don’t have any long-term software projects currently in progress, I have compared the new Gemini 3.8 Flash version with the free Gemini 3.6 Flash model on a variety of tasks: developing an interactive animation of the solar system, summarizing and simplifying a lengthy scientific paper, and analyzing the last week’s email in my Gmail account (one of the most challenging engineering tasks I know).
Let’s start with the fact that the Gemini 3.8 Flash was noticeably faster: 23 seconds versus 27 seconds to create a space project simulation, 15 seconds versus 20 seconds to generate a scientific report, and roughly the same time for sending emails. When you add up these figures for large projects and multiple tasks, it becomes clear that the new model truly does get the job done faster.
In the Solar System test, the simulation results were quite similar—perhaps reflecting how quickly they’ve progressed and now find this task relatively easy. However, Gemini 3.8 Flash performed slightly better in terms of rendering quality, with the planets appearing noticeably more “realistic.” Regarding the science overview, both AI models performed well in the complex task. In some ways, Gemini 3.6 Flash provided a clearer overview—perhaps because it’s a simpler model—but I found the visualizations and diagrams produced by Gemini 3.8 Flash more useful (remember, the images in both cases are generated by the Nano Banana 2, but the information is provided by the base model).
In the email task, the choice was slim: both accurately summarized my inbox contents and highlighted the same messages requiring attention (messages that turned out to be unimportant, but that’s another story). I still prefer to do it myself, but these AI models are becoming increasingly reliable.
The speedups and performance improvements Google touted are clearly present, even in small tests like this one. Notably, creating mini web apps and exploring complex physics papers are now easy for both Flash and Pro versions, which might be worth considering if you’re considering a Google AI plan. (Of course, if you want to upgrade to Gemini 3.8 Flash, you’ll have to pay.)