Gemini 3 is finally out.
I do not want to do clickbait, but I still want to say it.
This is the real global top tier model.
It is rare to see a model this anticipated. Weeks before launch, rumors were everywhere, and the whole AI community was buzzing.
From benchmarks alone, it is basically sweeping the charts.
Across 20 benchmarks, it beats Claude Sonnet 4.5 and GPT-5.1 on 19 of them. The only one where it trails slightly is SWE-Bench Verified, a coding task.
So what magic did Anthropic use on Claude
But across 20 benchmarks, covering multimodal images, video, math, tool use, reasoning, code, and long context, it does not just do one thing. It is both broad and extremely strong.
On LM Arena, it ranks first on almost every major list.
Text, vision, web dev, code, math all rank first by a wide margin.
They also shipped an IDE alongside Gemini 3 called AntiGravity.
You can tell they are confident.
I had tested Gemini 3 twice during the early leak phase, and it was already impressive. This time I wanted to share a few different cases.
Hands-on tests
SVG
Static SVG tasks are easy for Gemini 3 now. You can get animated SVGs quickly.
This works well for educational snippets. For example, a washing machine animation:
The prompt I used was fairly complex, but it nailed it.
I also asked it to generate a 360 degree draggable MacBook Pro SVG animation.
It even drew the Apple logo and the MacBook Pro notch, plus the signature space gray finish. The details are wild.
Web design
The biggest improvement in Gemini 3 is front end web design quality, especially overall aesthetics.
Try a new version with not gradients and super super high end and beautiful, unique and modern design
This looks amazing.
Another example:
Creat an esthetic website for criminal minds characters.
A high end website from a single prompt is no longer a dream.
On the multimodal side, Gemini 3 is also very strong. For example, it is much easier to do UI replication.
I fed it a screenshot of the Bilibili homepage and asked it to recreate it.
Use HTML, CSS, JS to recreate the Bilibili homepage
Aside from the cover images, it reproduced almost everything.
Here is a cyber style knowledge card using Gemini 3 official docs as content.
1. Objective
Build a cyberpunk interactive infographic focused on user provided content. Emphasize dynamic visuals and luminous UI.
2. Core Theme and Style
Theme: AI, virtual city, future economy.
Style: flicker, sharp, glowing.
Key aesthetics: neon lines and grid matrix, layered glass UI.
Secondary: dynamic particles and gradient flow to suggest energy.
3. Layout and Structure
- Top area with logo and system status bar, time, battery, connection.
- Main area with three rows of cards, stats panel, comparison analysis, conclusion visualization.
- Each card has hover highlight and fade in animation.
- Footer includes a neon outlined CTA button and copyright.
4. Design Elements and Visual Details
- Background gradient from deep blue to purple, #05091a to #1a1245.
- Primary colors: electric blue #3b82f6, neon purple #a855f7, acid green #84cc16.
- Glow effects, inner shadows, glass blur with backdrop-filter.
- Fonts: Orbitron or Roboto Mono.
- Line animation with keyframes for breathing light.
- Minimal corner radius and strong geometry.
5. Final Delivery
Single file HTML with a small amount of embedded CSS animation. Focus on presenting user content and visual motion with a futuristic atmosphere.
Looks great, and it pulled out the key information.
Video understanding
My favorite Gemini capability is video understanding. It is ahead of other models.
The most direct test is to generate a knowledge card from a video.
Test video link: https://www.youtube.com/watch?v=emWgP_fr04k
What is impressive is that it not only extracts key content, but also identifies main characters and links back to the source video.
The context window is huge. This video is over 40 minutes, and Gemini handled it without issues.
Another common scenario is breaking down short drama episodes.
The prompt is long so I will not include it here. If you are interested, DM me.
The video here is episode 1 of the short drama Peach Blossom Horse, Please Take the Spear.
Below is Gemini 3 Pro output so you can see the level of detail:
[
{
"time_range": "00:00:00-00:00:36",
"time": "Day",
"location": "Battlefield",
"roles": [
"Song Xixi",
"Enemy troops"
],
"dialogs": [
{
"desc": [
"Smoke fills the battlefield, corpses and flames everywhere, a tense atmosphere.",
"Song Xixi in armor with blood on her face charges on a white horse, resolute.",
"Subtitle appears: daughter of Marquis of Zhenbei, Song Xixi.",
"She swings a spear to push back attackers and recalls her mother."
],
"sub_dialog": [
{
"role": null,
"text": "Xixi, your father and brothers died in battle. Your mother wants you to hide your merit and avoid becoming a general. Live as a noblewoman, find a good husband, marry, and live a stable life."
},
{
"role": "Song Xixi",
"text": "Mother, if I can bring back father and brothers alive, I will agree. I will marry well and live a stable life."
}
]
},
{
"desc": [
"Song Xixi rides out of the city gate with enemies in pursuit.",
"Title appears: Peach Blossom Horse, Please Take the Spear."
],
"sub_dialog": []
}
]
},
{
"time_range": "00:00:36-00:01:17",
"time": "Night",
"location": "Outside Wenxi Residence",
"roles": [
"Song Xixi",
"Zhan Beiwang"
],
"dialogs": [
{
"desc": [
"Night scene with a mansion decorated in red ribbons, plaque reads Wenxi Residence.",
"Song Xixi in a red wedding dress adjusts Zhan Beiwang armor before he leaves for war.",
"Zhan Beiwang holds her hand and looks at her."
],
"sub_dialog": [
{
"role": "Zhan Beiwang",
"text": "I love only Xixi. I will never take a concubine. When I return safely, I will stay with you for life."
}
]
},
{
"desc": [
"Song Xixi is tearful but steady.",
"Zhan Beiwang embraces her under the red ribbons."
],
"sub_dialog": [
{
"role": "Song Xixi",
"text": "Do not worry, husband. I will take care of mother and protect the general residence until you return."
}
]
}
]
},
{
"time_range": "00:01:17-00:02:06",
"time": "Day",
"location": "Outside Wenxi Residence",
"roles": [
"Song Xixi",
"Zhan Beiwang",
"Maid"
],
"dialogs": [
{
"desc": [
"Subtitle appears: the following year.",
"Song Xixi in light green stands at the gate with a maid.",
"Zhan Beiwang returns in armor, cold and distant."
],
"sub_dialog": [
{
"role": "Zhan Beiwang",
"text": "Xixi, I fell in love with Yi Fang on the battlefield and used my merit to ask for a marriage decree. The decree is granted. She will enter as an equal wife, with no rank above or below you."
}
]
},
{
"desc": [
"Song Xixi looks shocked and disappointed."
],
"sub_dialog": [
{
"role": "Song Xixi",
"text": "Do you remember you promised never to take a concubine?"
}
]
},
{
"desc": [
"Zhan Beiwang looks indifferent and impatient."
],
"sub_dialog": [
{
"role": "Zhan Beiwang",
"text": "Forget that. I did not understand love then. You were suitable as my wife. Then I met Ah Fang. She is different from every woman I have seen. I love her deeply."
},
{
"role": "Song Xixi",
"text": "I want to meet General Yi Fang."
}
]
}
]
}
]
I tested this case one year ago and the results were bad. It missed content or mismatched dialogue and characters, so it was not usable.
Today, Gemini 3 does it extremely well.
Closing thoughts
Honestly, I have never anticipated a model this much. Before, I did not know what I was waiting for. Was it a stronger model, or just benchmark hype
After testing the official release, I realized the expectation was beyond the model itself.
When the final bell rings in the temple of AI, the voice most likely to open the door to ASI may come from Google.
Thank you for reading.
If this helped, please like, share, and follow. Do not miss updates and consider starring the account.