Google Gemini 3 launched a surprise attack worldwide and dealt a heavy blow to GPT-5.1. Sam Altman rarely offered congratulations.

In the early morning, Google's ultimate weapon, Gemini 3, made a grand entrance. Right off the bat, it came with the top - tier Pro version, touted as the AI powerhouse that combines the "strongest reasoning in history, multi - modality, and ambient programming"! It dominated the benchmark tests, even outperforming GPT - 5.1, thus heralding the next era of AI.

Here it comes! Here it comes!

Just now, the long - awaited annual highlight, Google's new flagship, Gemini 3, made a stunning debut.

Moreover, it started with the top - spec Gemini 3 Pro —

The model with the strongest reasoning, the most powerful multi - modality understanding, and the best capabilities in "agents" and "ambient programming" to date!

How powerful is it?

Just one hour after its release, even Altman, the CEO of OpenAI, personally tweeted his congratulations!

Moreover, it was in a case - sensitive version. (I wonder if he tried it himself)

Judging from the actual tests, this is indeed the case.

In numerous benchmark tests, Gemini 3 Pro reigned supreme —

It not only achieved an all - around performance leap compared to 2.5 Pro but also left OpenAI's newly launched GPT - 5.1 far behind.

To sum it up in Google's words, the core strengths of Gemini 3 Pro lie in these three aspects —

It topped the LMArena (1501 points) and WebDev (1487 points) rankings

It scored a record - high 45.8% in the Human Last Exam (HLE), demonstrating doctor - level reasoning ability

It is the king in the long - range task planning test, Vending - Bench 2

Moreover, in the enhanced reasoning mode, Gemini 3 Deep Think scored 41% in HLE, 93.8% in GPQA, and 45.1% in ARC - AGI - 2.

This day is destined to be written into history. As soon as Gemini 3 appeared, the whole internet went wild.

Gemini 3 is opening the next era of AI. Are you ready to hop on board?

As of today, the preview version of Gemini 3 Pro will be fully launched.

The Deep Think mode will take some time before it is available to Google AI Ultra subscribers.

Three Key Points (Condensed Version)

The birth of Gemini 3 marks another significant step for Google on the path to AGI!

First of all, it has extremely strong thinking ability, can deeply understand problems, and provide more insightful answers.

In particular, it is especially good at answering various complex scientific questions.

Build, deconstruct, and recombine detailed 3D voxel art with code

Secondly, it has world - leading multi - modality understanding ability, handling text, videos, and code with ease.

For example, it can interpret long videos or turn research papers into interactive guides.

In ambient programming, Gemini 3 has reached new heights.

In a nutshell, it can create a beautiful and dynamic application with just a simple description. Moreover, it can accurately grasp the intention and know how to implement it.

Meanwhile, its agent coding ability has been further enhanced. It seamlessly integrates with existing tools and is a perfect match for the new platform, Google Antigravity.

Gemini 3 Pro

Doctor - Level Reasoning Crushes All

With its top - notch reasoning and multi - modality capabilities, Gemini 3 Pro can turn any idea into reality!

It completely outperforms its predecessor, 2.5 Pro, leading by a large margin in all core benchmark tests.

·It topped the LMArena leaderboard, scoring a breakthrough 1501 Elo points;

·In the Human Last Exam (HLE), it scored 37.5% without using any tools;

·It achieved a high score of 91.9% in GPQA Diamond, demonstrating doctor - level reasoning ability;

·It set a new SOTA record of 23.4% in MathArena Apex, establishing a new benchmark in the field of mathematics.

Gemini 3 leads by a wide margin in a series of key AI benchmark tests

In addition to its excellent performance in text tests, Gemini 3 Pro is also the king of multi - modality —

It scored an impressive 81% in MMMU - Pro and 87.6% in Video - MMMU, redefining multi - modality reasoning.

It also achieved an industry - leading score of 72.1% in SimpleQA Verified, showing great improvement in factual accuracy.

This means that Gemini 3 Pro has the high - reliability ability to tackle complex problems in many fields such as science and mathematics.

Every interaction with Gemini 3 Pro comes with unprecedented "depth and subtlety".

Its answers are smart, concise, and straightforward, avoiding clichés and flattery, and providing real insights — Telling you what you need to hear, not just what you want to hear.

It's like a true thinking partner, offering a new way to understand information and express oneself.

Whether it's generating high - fidelity visualization code, explaining obscure scientific concepts, or having a creative brainstorming session, Gemini 3 Pro can handle it all.

Gemini 3 can write visualization code for plasma flow in a tokamak device and compose a poem capturing the essence of fusion physics

On Google AI Studio, the API pricing for Gemini 3 Pro is as follows —

Gemini 3 Deep Think

A New Peak of Intelligence

This time, Gemini 3 Deep Think officially ushers in a new era of "deep thinking", expanding the boundaries of intelligence once again.

Based on the reasoning and multi - modality understanding abilities of Gemini 3, it has achieved a qualitative leap and is better at tackling complex problems.

In multiple benchmark tests, Gemini 3 Deep Think outperformed Gemini 3 Pro:

It scored 41% (without using tools) in HLE and 93.8% in GPQA Diamond.

Moreover, it set a new historical high of 45.1% in ARC - AGI - 2 (with code execution, ARC Prize Verified), demonstrating its strong ability to handle unknown and novel problems.

Gemini 3 Deep Think performs excellently in some of the most challenging AI benchmark tests

Reshape the World, A New Era Begins

It can be said that Gemini 3 officially kicks off a new round of full - modality revolution!

One million tokens, full - modality explosion

Since its inception, Gemini has been designed for "cross - multi - modality", spanning text, images, videos, audio, and code, and can freely navigate through various information forms.

Gemini 3 has achieved a breakthrough upgrade, integrating the most advanced reasoning, visual and spatial understanding, leading multi - language performance, and 1 million tokens of context.

It can help people learn in the way that suits them best.

Suppose you want to learn your family's traditional cooking methods. Gemini 3 can decipher and translate hand - written recipes in different languages