its newest LLM: Gemini 3. The mannequin is long-awaited and has been broadly mentioned earlier than its launch. On this article, I’ll cowl my first expertise with the mannequin and the way it differs from different frontier LLMs.
The aim of the article is to share my first impressions when utilizing Gemini 3, highlighting what works effectively and what doesn’t work effectively. I’ll spotlight my expertise utilizing Gemini 3 within the console and whereas coding with it.
Why it is best to use Gemini 3
For my part, Gemini 2.5 professional was already the perfect conversational LLM out there earlier than the discharge of Gemini 3. The one space I imagine one other LLM was higher at was Claude Sonnet 4.5 considering, for coding.
The rationale I imagine Gemini 2.5 professional is the perfect non-coding LLM is because of its:
- Skill to effectively discover the right info
- Low quantity of hallucinations
- Its willingness to disagree with me
I imagine the final level is a very powerful. Some individuals need heat LLMs that really feel good to speak to; nonetheless, I’d argue you (as a problem-solver) all the time need the other:
You need an LLM that goes straight to the purpose and is prepared to say that you’re incorrect
My expertise was that Gemini 2.5 was much better at this, in comparison with different LLMs akin to GPT-5, Grok 4, and Claude Sonnet 4.5.
Contemplating Google, for my part, already had the perfect LLM on the market, the discharge of a more moderen Gemini mannequin is thus very fascinating, and one thing I began testing proper after launch.
It’s price mentioning that Google launched Gemini 3 Professional, however has not but launched a flash different, although it’s pure to suppose such a mannequin will likely be launched quickly.
I’m not endorsed by Google within the writing of this text.
Gemini 3 within the console
I first began testing Gemini 3 Professional within the console. The very first thing that struck me was that it was comparatively sluggish in comparison with Gemini 2.5 Professional. Nonetheless, that is often not a difficulty, as I principally worth intelligence over velocity, after all, as much as a sure threshold. Although Gemini 3 Professional is slower, I positively wouldn’t say it’s too sluggish.
One other level I seen is that when explaining, Gemini 3 creates or utilises extra pictures in its explanations. For instance, when discussing EPC certificates with Gemini, the mannequin discovered the picture beneath:
I additionally seen it could generally generate photos, even when I didn’t explicitly immediate for it. The picture technology within the Gemini console is surprisingly quick.
The second I used to be most impressed by Gemini 3’s capabilities was after I was analyzing the primary analysis paper on diffusion fashions, and I mentioned with Gemini to grasp the paper. The mannequin was, after all, good at studying the paper, together with textual content, photos, and equations; nonetheless, that is additionally a functionality the opposite frontier fashions possess. I used to be most impressed after I was chatting with Gemini 3 about diffusion fashions, attempting to grasp them.
I made a false impression concerning the paper, considering we had been discussing conditional diffusion fashions, although we had been in truth taking a look at unconditional diffusion. Word that I used to be discussing this earlier than even realizing concerning the phrases conditional and unconditional diffusion.
Gemini 3 then proceeded to name out that I used to be misunderstanding the ideas, effectively understanding the actual intent behind my query, and considerably helped me deepen my understanding of diffusion fashions.
I additionally took a few of the older queries I ran within the Gemini console with Gemini 2.5 Professional, and ran the very same queries once more, this time utilizing Gemini 3 Professional. They had been often broader questions, although not significantly troublesome ones.
The responses I acquired had been general fairly comparable, although I did discover Gemini 3 was higher at telling me issues I didn’t know, or uncovering matters / areas I (or Gemini 2.5 Professional) hadn’t considered earlier than. I used to be, for instance, discussing how I write articles, and what I can do to enhance, the place I imagine Gemini 3 was higher at offering suggestions, and arising with extra inventive approaches to bettering my writing.
Thus, to sum it up, Gemini 3 within the console is:
- A bit sluggish
- Good, and gives good explanations
- Good at uncovering issues I haven’t considered, which is tremendous useful when coping with problem-solving
- Is prepared to disagree with you, and assist name out ambiguities, traits I imagine are actually essential in an LLM assistant
Coding with Gemini 3
After working with Gemini 3 within the console, I began coding with it by Cursor. My general expertise is that it’s positively a great mannequin, although I nonetheless choose Claude Sonnet 4.5 considering as my fundamental coding mannequin. The primary purpose for that is that Gemini 3 too typically comes up with extra advanced options and is a slower mannequin. Nonetheless, Gemini 3 is most positively a really succesful coding mannequin that is perhaps higher for different coding use-cases than what I’m coping with. I’m principally coding infrastructure round AI brokers and CDK stacks.
I attempted Gemini 3 for coding in two fundamental methods:
- Making the sport proven on this X put up, from only a screenshot of the sport
- Coding some agentic infrastructure
First, I tried to make the Sport from the X put up. On the primary immediate, the mannequin made a Pygame with all of the squares, however it forgot so as to add all of the sprites (artwork), the bar on the left aspect, and so forth. Principally, it made a really minimalist model of the sport.
I then wrote a follow-up immediate with the next:
Make it look correctly like this recreation with the design and every thing. UseWord: When coding, try to be far more particular in your directions than my immediate above. I used this immediate as a result of I used to be basically vibing within the recreation, and needed to see Gemini 3 Professional’s capacity to create a recreation from scratch.
After operating the immediate above, it made a working recreation, the place the visitors are strolling round, I can purchase pavements and totally different machines, and the sport basically works as anticipated. Very spectacular!
I continued coding with Gemini 3, however this time on a extra production-grade code base. My general conclusion is that Gemini 3 Professional often will get the job performed, although I extra typically expertise bloated or worse code than I do when utilizing Claude 4.5 Sonnet. Moreover, Claude Sonnet 4.5 is sort of a bit sooner, making it the particular mannequin of selection for me when coding. Nonetheless, I’d most likely regard Gemini 3 Professional because the second-best coding mannequin I’ve used.
I additionally suppose that which coding mannequin is finest extremely depends upon what you’re coding. In some conditions, velocity is extra essential. Specifically types of coding, one other mannequin is perhaps higher, and so forth, so it is best to actually check out the fashions your self and see what works finest for you. The value of utilizing these fashions goes down quickly, and you’ll simply revert any adjustments made, making it tremendous low-cost to check out totally different fashions.
It’s additionally price mentioning that Google launched a brand new IDE referred to as Antigravity, although I haven’t tried it but.
General impressions
My general impression of Gemini 3 is nice, and my up to date LLM utilization stack will appear like this:
- Claude 4.5 Sonnet considering for coding
- GPT-5 after I want fast solutions to easy questions (the GPT-app works effectively to open with a shortcut).
- GPT-5 when producing photos
- Once I need extra thorough solutions and have longer discussions with an LLM a few subject, I’ll use Gemini 3. Sometimes, to study new matters, talk about software program structure, or comparable.
The pricing for Gemini 3 per million tokens appears to be like like the next (per November 19, 2025, from Gemini Developer API Docs)
- In case you have lower than 200k enter tokens:
- Enter tokens: 2 USD
- Output tokens: 12 USD
- In case you have greater than 200k enter tokens:
- Enter tokens: 4 USD
- Output tokens: 18 USD
In conclusion, I’ve good first impressions from Gemini 3, and extremely suggest checking it out.
👉 Discover me on socials:
💻 My webinar on Imaginative and prescient Language Fashions
📩 Subscribe to my e-newsletter
🧑💻 Get in contact
✍️ Medium
You too can learn my different articles:
