Hey everyone, I'm Achao! Today I'm here to share some major news—Google just unveiled their most powerful AI model yet: Gemini 3 Pro! This isn't just any ordinary upgrade—it's genuinely ”scary powerful.”
Introduction: What is Gemini 3 Pro?
Gemini 3 Pro is Google's latest large language model, officially described as ”our smartest model yet.” Simply put, it's like aPhD-level AI assistantNot only are they well-versed in knowledge, but more importantly, they possess the ability to think deeply and solve problems independently.
Let me break it down in plain English: If previous AI models were like knowledgeable librarians, then Gemini 3 Pro is a PhD-level assistant who can think deeply, grasp abstract requests, and take proactive steps to solve problems.
Key Feature Highlights
1. Unprecedented reasoning ability
Gemini 3 Pro has achieved a quantum leap in reasoning capabilities. It delivers remarkable performance across all benchmarks:
- LMArena Rankings: Topping the charts with an Elo rating of 1501, surpassing all competitors
- Humanity's Final TestAchieving a score of 37.51 TP3T without using tools demonstrates doctoral-level reasoning skills.
- Mathematical abilityAchieved a new high of 23.41 TP3T on MathArena Apex
The most impressive thing about it is itsDeep Think ModeThis is akin to AI's ”slow thinking” mode. When confronted with complex problems, it doesn't rush to respond. Instead, it first conducts multi-step internal reasoning, self-contradiction, and planning to ensure the accuracy of its answers.
2. Robust multimodal understanding
Gemini 3 Pro excels at understanding multiple media formats:
- Text, images, PDFs, videosComprehensive analysis
- be in favor of1 million tokensExtremely long content input—capable of reading an entire book or an entire codebase.
- Capable of understanding handwritten recipes, analyzing academic papers, and even analyzing sports game footage.
For example, you can upload your family's handwritten recipes, and it will transcribe and compile them into a shareable family cookbook; or upload your pickleball match videos, and it will identify areas for improvement and create a training plan.
3. Revolutionary Proxy Capabilities
This is the most exciting feature of Gemini 3 Pro! It no longer just offers suggestions—it canActually help you complete the task::
- Automatic Planning Steps: Capable of independently planning complex workflows
- Tool Usage: Can invoke tools such as browsers and terminals to execute tasks.
- Long-term planningDemonstrated strong long-term planning capabilities in the Vending-Bench 2 test.
For example, you can authorize it to organize your Gmail inbox or plan your travel itinerary, and it will guide you step by step through these complex tasks.
4. Exceptional programming skills
For developers and engineers, the Gemini 3 Pro is an absolute game-changer:
- Vibe Coding: Capable of interpreting ambiguous design concepts and generating complete code based on simple descriptions.
- Code Fix: Scored 76.21 TP3T on SWE-bench Verified, capable of automatically fixing real GitHub issues
- Terminal operationScored 54.21 TP3T in Terminal-Bench 2.0, capable of executing Unix environment commands.
You can describe ”I want a Cyberpunk-style 3D dashboard,” and it will generate complete code including HTML, CSS, React, and even WebGL.
Cost Information
The good news is that currentlyGemini 3 Pro is available for free use in Google AI Studio.General chatting, uploading pictures, or testing features are all free of charge, making it perfect for users who want to try it out first.
If you choose the API route, billing will be based on tokens (character count):
- Input (<200,000 tokens): Approximately $2 per 1 million tokens
- Output (<200,000 tokens): Approximately $12 per 1 million tokens
Gemini3Pro Official Website
- ￥Download for freeDownload after commentDownload after login
- {{attr.name}}:
