Author: Miao Zheng
After pretending to be asleep for eight months, Google has suddenly dropped a bombshell: the Gemini 3 Pro.
Google has finally released the Gemini 3 Pro, quite suddenly and in a very "low-key" manner.
Although Google released the Nano Banana image editing model before the Gemini 3 Pro, thus making a splash, it has been too silent on the issue of pedestal models for far too long.
Over the past six months, everyone has been discussing OpenAI's new moves or marveling at Claude's dominance in the coding field, but no one has mentioned Gemini, which hasn't had a version number update in eight months.
Even with its impressive cloud business and financial reports, Google's presence within the core circle of AI developers is gradually being diluted.
Fortunately, after experiencing it firsthand, we found that the Gemini 3 Pro did not disappoint.
However, it's too early to draw conclusions. The AI field has long since moved beyond the stage of intimidating users with sheer numbers of parameters; now everyone is focused on applications, deployment, and cost reduction.
Whether Google can adapt to the new version and the new environment remains to be seen.
01
I asked the Gemini 3 Pro to describe itself in one sentence, and this is how it answered me.
"Instead of rushing to prove how smart I am to the world, I'm starting to think about how to make myself more useful." — Gemini 3 Pro
On the LMArena leaderboard, the Gemini 3 Pro topped the list with an Elo score of 1501, setting a new record for AI models in comprehensive capability assessments. This is an outstanding achievement, and even Ultraman tweeted his congratulations.
In math aptitude tests, the model achieved 100% accuracy in the AIME 2025 (American Invitational Mathematics Competition) code execution mode. In the GPQA Diamond science knowledge test, the Gemini 3 Pro achieved an accuracy rate of 91.9%.
Results from the MathArenaApex math competition show that Gemini 3 Pro achieved a score of 23.4%, while other mainstream models generally scored below 2%. Furthermore, in the Humanity's LastExam test, the model achieved a score of 37.5% without using the tool.
In this update, Google introduced a code generation feature called "vibecoding." This feature allows users to describe their needs in natural language, and the system then generates the corresponding code and applications.
In tests conducted in the Canvas programming environment, after a user described "creating an electric fan with adjustable speed," the system generated complete code, including rotation animation, speed control slider, and on/off button, within approximately 30 seconds.
The official case studies also included visual simulations of the nuclear fusion process.
In terms of interaction, the Gemini 3 Pro adds a "Generative UI" feature. Unlike traditional AI assistants that only return text answers, this system can automatically generate a customized interface layout based on the query content.
For example, when a user asks questions related to quantum computing, the system may generate an interactive interface that includes explanations of concepts, dynamic diagrams, and links to relevant papers.
The system generates different interface designs for the same question to different audiences. For example, when explaining the same concept to children and adults, different presentation methods will be used. Children's interfaces will be more cute, while adults' interfaces will be more concise and clear.
The Visual Layout experimental feature provided by Google Labs demonstrates the application of this interface, allowing users to obtain a magazine-style view layout that includes images, modules, and adjustable UI elements.
This release also includes an intelligent agent system called Gemini Agent, which is currently in the experimental stage. This system can perform multi-step tasks and connect to Google services such as Gmail, Google Calendar, and Reminders.
In inbox management, the system can automatically filter emails, prioritize them, and draft replies. Travel planning is another application; users simply provide their destination and approximate dates, and the system queries the calendar, searches for flight and hotel options, and adds the itinerary. This feature is currently only available to Google AI Ultra subscribers in the United States.
In terms of multimodal processing, Gemini 3 Pro is built on a sparse hybrid expert architecture, supporting text, image, audio, and video inputs. The model's context window is 1 million tokens, meaning it can handle long documents or video content.
Tests conducted by Mark Humphries, a history professor at Laurier University in Canada, showed that the model had a character error rate of 0.56% when recognizing 18th-century handwritten manuscripts, a reduction of 50% to 70% compared to previous versions.
Google stated that the training data included publicly available web documents, code, images, audio, and video content, and reinforcement learning techniques were used in the post-training phase.
Google has also launched an optimized version called Gemini 3 Deep Think, specifically designed for complex inference tasks. This mode is currently undergoing security evaluation and is planned to be rolled out to Google AI Ultra subscribers in the coming weeks.
In Google Search's AI mode, users can click the "thinking" tab to view the reasoning process. Compared to the standard mode, Deep Think mode performs more steps of analysis before generating an answer.
In addition to the official information, I also compared the Gemini 3 Pro with ChatGPT-5.1.
The first comparison is the generated image.
Prompt: Generate an iPhone 17 image for me
ChatGPT-5.1
Gemini 3 Pro
Subjectively speaking, ChatGPT-5.1 better meets my needs, so ChatGPT-5.1 wins this round.
The second comparison is between the two in terms of their agent level.
Hint: Go and research the WeChat public account "Alphabet Ranking" for me, and then comment on the quality of this account.
GPT-5.1
Gemini 3 Pro
While I personally prefer the interpretation of Gemini 3 Pro, it's too over-hyped. ChatGPT-5.1 reveals some shortcomings in the rankings and is more objective and truthful.
Finally, there's coding ability, which is currently the most important aspect for all large-scale models.
The project I chose is LightRAG, a highly-rated project on GitHub recently. It improves retrieval by integrating graph structures to enhance context awareness and efficient information retrieval, resulting in higher accuracy and faster response times. Project address: https://github.com/HKUDS/LightRAG
Prompt: Tell me about this project.
GPT-5.1
Gemini 3 Pro
Meanwhile, the Gemini 3 Pro has also received high praise from industry professionals.
02
Although the Gemini 3 Pro was released very quietly, Google had actually been teasing it for a long time.
During Google's third-quarter earnings call, Google CEO Sundar Pichai said, "The Gemini 3 Pro will be released sometime in 2025." Without a specific date or further details, he kicked off a major marketing saga in the tech industry.
Google has been sending signals to keep the entire AI community on high alert, but has consistently refused to give any definite release timeline.
Starting in October, a series of "accidental leaks" began to surface. On October 23, a calendar began circulating, with screenshots of an internal calendar for the "Gemini 3 Pro Release" on November 12 going viral.
Furthermore, sharp-eyed developers also discovered the phrase "gemini-3-pro-preview-11-2025" in Vertex AI's API documentation.
Following this, various screenshots began appearing on Reddit and X. Some users claimed to have seen the new model in the Gemini Canvas tool, while others discovered unusual model identifiers in certain versions of the mobile app.
Then, the following test data began to circulate on social media.
These "leaks" may seem accidental, but they actually constitute a carefully orchestrated prelude.
Each leak perfectly showcases a core capability of the Gemini 3 Pro, and each discussion pushes expectations to new heights. Google's official account, however, takes an intriguing stance. They retweet community discussions, use phrases like "coming soon" to whet the appetite, and even senior executives at Google AI Labs replied with two "thinking" emojis to a tweet about a predicted release date, but they refuse to give a precise date.
After nearly a month of anticipation, Google has finally unveiled the new Gemini 3 Pro. However, while the Gemini 3 Pro boasts powerful performance, Google's update frequency is somewhat frustrating.
Back in March of this year, Google released a preview version of Gemini 2.5 Pro, followed by derivative preview versions such as Gemini 2.5 Flash. Until the release of Gemini 3 Pro, the Gemini series did not receive any version number upgrades during this period.
But Google's competitors won't wait for Gemini.
OpenAI launched GPT-5 on August 7th and further upgraded it to GPT-5.1 on November 12th. During this period, OpenAI also launched its own AI browser, Atlas, directly targeting Google's core market.
Anthropic's iteration speed is even more frequent: Claude 3.7 Sonnet (the first hybrid inference model) was released on February 24, Claude Opus 4 and Sonnet 4 were released on May 22, Claude Opus 4.1 was released on August 5, Claude Sonnet 4.5 was released on September 29, and Claude Haiku 4.5 was released on October 15.
This series of attacks caught Google somewhat off guard, but so far, it seems Google has withstood the pressure.
03
The biggest reason why Google took eight months to update the Gemini 3 Pro is probably due to personnel changes.
Around July to August 2025, Microsoft launched a fierce talent offensive against Google, successfully recruiting more than 20 core experts and executives from DeepMind.
This includes Dave Citron, Senior Director of Product at DeepMind, who is responsible for the implementation of its core AI products, and Amar Subramanya, VP of Engineering at Gemini, one of the core engineering leaders for Google's most important model, Gemini.
On the other hand, the Google Nano Banana team stated that Google struggled with the field of AI-generated images for a long time after the release of Gemini 2.5 Pro, thus slowing down the updates of the base model.
Google believes that only by overcoming the three major challenges in the field of image generation—character consistency, in-context editing, and text rendering—can the pedestal model perform better.
The Nano Banana team stated that the model can not only "draw beautifully," but more importantly, it can "understand human language" and be "controlled by humans," thus enabling AI-generated images to truly enter the commercial application stage.
Looking back at the Gemini 3 Pro now, it's a passable answer, but in this fast-paced AI battlefield, simply getting by is no longer enough.
Since Google has chosen to submit its report at this moment, it must be prepared to face the most demanding examiners—users and developers whose tastes have been spoiled by competitors. The next few months will not be a competition of model parameters, but a brutal battle of ecosystem integration capabilities. Google, this elephant, not only needs to learn to dance, but it also needs to dance faster than everyone else.
