Name: 2026年你絕對需要的 Google Gemini 3.0 Pro 功能全解析！ (Every Google Gemini 3.0 Pro Feature You'll NEED in 2026)
Uploaded: 2026-01-15T00:29:31.000Z
Duration: 13 min 35 s
Description: 【看影片學英語】數萬部 YouTube 影片，搭配英漢字典即點即查，輕鬆掌握單字發音與用法，長久累積看電影不必再看字幕。

程度副詞

And if you're still using it like a basic search engine, you're missing out on features that could literally save you hours every single day.

如果你還只是把它當成基本搜尋引擎在用，那你就錯過了一堆能幫你省下超多時間的功能。

情態副詞

Most people have no idea what this thing can actually do.

大部分人都不知道它到底能做些什麼。

They type a question, get an answer and move on.

他們只會問個問題，得到答案就結束了。

But Gemini 3.0 Pro has tools buried inside that can analyze entire videos, generate studio quality images, build interactive apps, and

但 Gemini 3.0 Pro 裡面藏著很多工具，可以分析整支影片、生成錄音室等級的圖片、建立互動式應用程式，

even create full podcasts from your documents.

甚至能根據你的文件，製作出完整的Podcast。

And in this video, I'm going to show you every single one of them.

在這支影片裡，我會把每一個功能都教給大家。

Right now think of it as the brain powering all of Google's AI tools.

現在你可以把它想像成驅動 Google 所有 AI 工具的大腦。

It's multimodal, which means it doesn't just read text, it can understand and generate text, images, audio, video, PDFs and even entire code bases all at once.

它是多模態的，這表示它不只讀文字，還能同時理解和生成文字、圖片、音訊、影片、PDF 甚至完整的程式碼庫。

Gemini 3.0 Pro launched in November 2025 and it's a massive upgrade from the previous version.

Gemini 3.0 Pro 在 2025 年 11 月推出，相較於前一代是個巨大的升級。

The model is designed for complex reasoning, long context understanding and what they call agentic tasks, which basically means it can plan, execute and complete workflows with multiple steps, surpassing any other AI model that would struggle to keep up with all the information.

這個模型專為複雜推理、長篇內容理解以及所謂的代理任務而設計，基本上意味著它能規劃、執行和完成多步驟的工作流程，超越任何難以跟上所有資訊的其他 AI 模型。

Now, if you care about the numbers, it scores 37.5% on humanity's last exam, which is a benchmark for advanced reasoning.

現在，如果你在意數字，它在 humanity's last exam 上拿到 37.5%，這是進階推理的基準。

It hits 91.9% on GPQA diamond, which tests graduate level knowledge, and

它在 GPQA diamond 上達到 91.9%，這個測試的是研究所等級的知識，

72.7% on ScreenSpot Pro, which measures how well it understands user interface elements on a screen.

以及 72.7% 在 ScreenSpot Pro 上，這個衡量它對螢幕上使用者介面元素的理解程度。

It outperforms GPT 5.1, Claude Sonnet 4.5 and its own predecessor across the board.

它在各方面都超越了 GPT 5.1、Claude Sonnet 4.5 以及它自己的前身。

Alright, let me show you how this actually works.

好的，我來示範一下實際運作。

I'm logging into Gemini right now and we're going to run a few tests.

我現在就登入 Gemini，我們要來跑幾個測試。

I want to show you the difference between low thinking level and high thinking level.

我想讓大家看看低思考層級跟高思考層級的差別。

And then we'll test the multimodal capabilities.

接著我們會測試它的多模態能力。

I've selected thinking from the model picker.

我在模型選擇器裡選了「思考」。

這樣我就可以使用 Gemini 3.0 Pro 了。

我們就從一個簡單的測試開始吧。

Option B gives you 12% return but a $200 yearly fee.

選項 B 每年給你 12% 的報酬，但要付 200 美元的年費。

Which one makes more money after five years?

Watch how Gemini 3.0 Pro thinks through this.

來看看 Gemini 3.0 Pro 是怎麼思考這個問題的。

It's doing the math, figuring out the fees, and comparing both options side by side.

它正在計算、考量費用，並比較這兩種選項。

It's not just picking one randomly, it's showing me the actual numbers for both choices and telling me which one wins.

它不是隨便選一個，而是給我看兩個選項的實際數字，並告訴我哪個比較好。

The model took a few extra seconds to think before answering, and

這個模型花了幾秒鐘思考後才回答，

the result is way cleaner and easier to understand.

而且結果更清晰、更容易理解。

Now let's test the multimodal capabilities.

現在我們要來測試它的多模態能力。

I'm going to upload an image, a screenshot of a complex data dashboard and ask Gemini to analyze it.

我要上傳一張圖片，一個複雜的數據儀錶板截圖，然後請 Gemini 分析它。

And as you can see, it's not just reading the image, it's actually interpreting the data, spotting patterns, calling out anomalies, and

而且就像你看到的，它不只是讀取圖片，它還在解讀數據，找出模式，點出異常，並且

This is the moment you realize multimodal models aren't just a small change, they're basically a full analyst who can look at a dashboard and tell you what matters before you've even zoomed in.

這就是你會意識到多模態模型不只是一個小小的改變，牠們基本上就是一個完整的分析師，可以看著儀錶板，在你還沒放大之前就告訴你什麼是重要的。

Let's also test its video analysis features.

我們也來測試一下它的影片分析功能。

In the next example, I'll upload this video and ask Gemini to break it into chapters, highlight the key moments, identify emotions, and

在下一個例子裡，我會上傳這段影片，然後請 Gemini 將它抽成章節，標出關鍵時刻，辨識情緒，並且

The goal is can a multimodal model act like a real editor and not just a transcription machine?

目標是，一個多模態模型能不能像真正的編輯一樣運作，而不只是個轉錄機器？

And as you can see, it does the whole thing.

而且就像你看到的，它把所有事情都做好了。

Clean structure, clear insights, even notes on pacing.

結構清晰，見解明確，甚至還有關於節奏的筆記。

At this point, Gemini isn't just helping with the editing, but it completely structures it.

這時候，Gemini 不只是協助編輯，而是完全構建了它。

The final test is for long context understanding.

最後的測試是關於長篇內容的理解能力。

I'm going to upload a 30 page PDF, a research report, and

我會上傳一份 30 頁的 PDF，一份研究報告，然後

ask Gemini to summarize it and answer specific questions.

請 Gemini 幫我總結並回答特定問題。

Pulled out the key findings, identified the methodology and flagged the limitations section.

抓出了關鍵發現，辨識了研究方法，並標示了限制的部分。

This is the 2 million token context window at work.

這就是 200 萬 token 上下文窗口的威力。

While other tools would struggle with this much information and sometimes even ignore most of it, Gemini has the capacity to understand and memorize almost anything you give it.

其他工具在處理這麼多資訊時會很吃力，有時甚至會忽略大部分內容，但 Gemini 卻能理解並記住你給它的幾乎所有東西。

Now let's test the new extended AI mode in Google Search.

我們來測試一下 Google 搜尋裡新的擴充式 AI 模式。

this is Gemini 3.0 Pro powering Google Search directly, giving you deeper, more context aware answers.

這是 Gemini 3.0 Pro 直接驅動 Google 搜尋，給你更深入、更具情境意識的答案。

I've enabled AI mode with extended thinking in Google Search.

我在 Google 搜尋裡啟用了具備擴充式思考的 AI 模式。

I type in best budget laptops for students in 2025 and look at what it returns.

我輸入「2025 年最適合學生的預算型筆記型電腦」，看看它會回傳什麼。

它不只是一堆藍色連結的頁面。

It gives me a curated set of recommended picks with quick specs, pricing and source callouts plus a clean key considerations Battery portability, performance, storage, durability.

它提供我一套精選的推薦機種，包含快速規格、價格和來源標示，還有清楚的重點考量：電池續航、攜帶性、效能、儲存空間、耐用度。

It even surfaces shopping cards with prices and retailers right there in the layout.

它甚至直接在版面裡顯示了包含價格和零售商的購物卡。

現在我們來試試更複雜一點的。

Gemini doesn't just give surface level text, it explains that entanglement links particles into a single shared system so their properties aren't fully defined until you measure them.

Gemini 不只會給表面上的文字，它會解釋糾纏如何將粒子連結成一個單一共享系統，所以在你測量它們之前，它們的屬性並不會完全定義。

Then it breaks down the key principles like superposition, instantaneous correlation, and

然後它會分解關鍵原則，像是疊加態、瞬間關聯性，以及

why it looks non local without letting you send messages faster than light.

為什麼它看起來是非局域的，同時又不會讓你以超光速傳送訊息。

It even clarifies what it isn't and uses an analogy to make the difference.

它甚至澄清了它不是什麼，並使用類比來區分。

Click this is AI learning at its best, turning a confusing physics idea into something clear and easy to understand.

這就是 AI 學習的極致，將一個令人困惑的物理概念轉化為清晰易懂的東西。

I ask Visualize how compound interest grows over 20 years with $10,000 at a 7% annual return.

我問「視覺化 20 年來，以每年 7% 的報酬率，10,000 美元的複利如何增長。」

It shows the growth curve, explains why it accelerates over time, gives a clear year by year breakdown, and

它會顯示成長曲線，解釋為什麼會隨著時間加速，提供清楚的逐年細項，並且

calculates the final value after 20 years.

So we've seen AI mode in search, but Gemini 3.0 Pro becomes seriously powerful for education.

所以我們看過了搜尋中的 AI 模式，但 Gemini 3.0 Pro 在教育方面變得超級強大。

It actually markets specifically to college students by offering a free full year of use on the paid plan.

它實際上是針對大學生行銷，提供付費方案免費使用一整年。

You can ask it to make learning materials, turn hard ideas into visuals, and

你可以請它製作學習教材，將難懂的概念視覺化，並且

even build simple interactive tools to explore science.

甚至建立簡單的互動工具來探索科學。

It's not just about reading explanations anymore.

You can create visual hands on learning experiences whenever you need them.

你可以隨時隨地創造你需要的視覺化動手做學習體驗。

I'm going to ask Gemini to make a visual explanation of a physics idea.

我要請 Gemini 製作一個物理概念的視覺化解釋。

It shows the curved path of the object, breaks the velocity into the horizontal part that stays constant and the vertical part that changes because of gravity and explains what's happening across the motion in clear stages.

它顯示了物體的曲線路徑，將速度分解為保持恆定的水準部分和因重力而改變的垂直部分，並清楚地分階段解釋了運動過程中發生的情況。

It even includes simple diagrams for the launch angle, the highest point where the vertical speed hits zero, and

它甚至包含了發射角度、垂直速度為零的最高點，以及

where it finally lands with the velocity components shown.

Now I'm going to ask Gemini to write code that simulates this coding.

現在我要請 Gemini 寫一段程式碼來模擬這個。