3 Pro
Best for complex tasks and bringing creative concepts to life
Introducing our most intelligent model yet. With state-of-the-art reasoning to help you learn, build, and plan anything.
3 Flash
Best for frontier intelligence at speed
2.5 Flash-Lite
Best for high volume, cost efficient tasks
Gemini 1 introduced native multimodality and long context to help AI understand the world. Gemini 2 added thinking, reasoning and tool use to create a foundation for agents. Now, Gemini 3 brings these capabilities together – so you can bring any idea to life.
Slide 1 of 3
Learn anything
Understand complex topics in a way that makes sense for you – with clear, concise, and helpful responses
Build anything
Bring your ideas to life – from sketches and prompts to interactive tools and experiences
Plan anything
Delegate tasks and multi-step projects to get things done faster than ever before
Build with Google Antigravity – our AI-first developer experience
Slide 1 of 6
Created with Gemini 3 Pro
Code a 3D visualization of the universe
Gemini 3 uses state-of-the-art reasoning to generate richer visualizations and deeper interactivity. See how it codes a seamless 3D journey through the scale of the universe, from a proton to the observable universe, demonstrating a massive leap in "vibe coding" performance over Gemini 2.5.
Created with Gemini 3 Flash
Visual context in an instant
Leverage Gemini 3 Flash’s multimodal capabilities in visual recognition and reasoning to add contextual UI on image generations. 3 Flash has the capability to describe the content of the image in a compelling and interactive way.
Created with Gemini 3 Pro
Interact with complex topics like RNA transcription
Gemini 3’s state-of-the-art reasoning provides unprecedented nuance and depth
Created with Gemini 3 Flash
Assist in near real-time game play
In this slingshot game, Gemini 3 Flash delivers near real-time strategic guidance by simultaneously analyzing the video and hand-tracking inputs. It handles complex geometric calculations and velocity estimation to enable responsive live assistance.
Created with Gemini 3 Pro
Turn treasured recipes into shareable family cookbooks
Gemini 3 seamlessly synthesizes information across text, images, video, audio, and even code to help you learn. Generate code for interactive flashcards, games and experiences to help you master new material.
Created with Gemini 3 Flash
Creative UI in a spark
Generate new UIs instantly with Gemini 3 Flash, explore multiple creative variations, and interact with 3 Flash in near real-time to have it come up with best UI outcomes, all with one click.
| Benchmark | Notes | Gemini 3 Flash Thinking | Gemini 3 Pro Thinking | Gemini 2.5 Flash Thinking | Gemini 2.5 Pro Thinking | Claude Sonnet 4.5 Thinking | GPT-5.2 Extra high | Grok 4.1 Fast Reasoning |
|---|---|---|---|---|---|---|---|---|
| Input price | $/1M tokens | $0.50 | $2.00 $4.00 > 200k tokens | $0.30 | $1.25 $2.50 > 200k tokens | $3.00 $6.00 /MTok > 200k tokens | $1.75 | $0.20 |
| Output price | $/1M tokens | $3.00 | $12.00 $18.00 > 200k tokens | $2.50 | $10.00 $15.00 > 200k tokens | $15.00 $22.50 > 200k tokens | $14.00 | $0.50 |
| Academic reasoning (full set, text + MM) Humanity's Last Exam | No tools | 33.7% | 37.5% | 11.0% | 21.6% | 13.7% | 34.5% | 17.6% |
| With search and code execution | 43.5% | 45.8% | — | — | — | 45.5% | — | |
| Visual reasoning puzzles ARC-AGI-2 | ARC Prize Verified | 33.6% | 31.1% | 2.5% | 4.9% | 13.6% | 52.9% | — |
| Scientific knowledge GPQA Diamond | No tools | 90.4% | 91.9% | 82.8% | 86.4% | 83.4% | 92.4% | 84.3% |
| Mathematics AIME 2025 | No tools | 95.2% | 95.0% | 72.0% | 88.0% | 87.0% | 100% | 91.9% |
| With code execution | 99.7% | 100% | 75.7% | — | 100% | — | — | |
| Multimodal understanding and reasoning MMMU-Pro | 81.2% | 81.0% | 66.7% | 68.0% | 68.0% | 79.5% | 63.0% | |
| Screen understanding ScreenSpot-Pro | No tools unless specified | 69.1% | 72.7% | 3.9% | 11.4% | 36.2% | 86.3% with python | — |
| Information synthesis from complex charts CharXiv Reasoning | No tools | 80.3% | 81.4% | 63.7% | 69.6% | 68.5% | 82.1% | — |
| OCR OmniDocBench 1.5 | Overall Edit Distance, lower is better | 0.121 | 0.115 | 0.154 | 0.145 | 0.145 | 0.143 | — |
| Knowledge acquisition from videos Video-MMMU | 86.9% | 87.6% | 79.2% | 83.6% | 77.8% | 85.9% | — | |
| Competitive coding problems from Codeforces, ICPC, and IOI LiveCodeBench Pro | Elo Rating, higher is better | 2316 | 2439 | 1143 | 1775 | 1418 | 2393 | — |
| Agentic terminal coding Terminal-Bench 2.0 | Terminus-2 harness | 47.6% | 54.2% | 16.9% | 32.6% | 42.8% | — | — |
| Agentic coding SWE-bench Verified | Single attempt | 78.0% | 76.2% | 60.4% | 59.6% | 77.2% | 80.0% | 50.6% |
| Agentic tool use τ2-bench | 90.2% | 90.7% | 79.5% | 77.8% | 87.2% | — | — | |
| Long horizon real-world software tasks Toolathlon | 49.4% | 36.4% | 3.7% | 10.5% | 38.9% | 46.3% | — | |
| Multi-step workflows using MCP MCP Atlas | 57.4% | 54.1% | 3.4% | 8.8% | 43.8% | 60.6% | — | |
| Agentic long term coherence Vending-Bench 2 | Net worth (mean), higher is better | $3,635 | $5,478 | $549 | $574 | $3,839 | $3,952 | $1,107 |
| Factuality benchmark across grounding, parametric, search, and MM FACTS Benchmark Suite | 61.9% | 70.5% | 50.4% | 63.4% | 48.9% | 61.4% | 42.1% | |
| Parametric knowledge SimpleQA Verified | 68.7% | 72.1% | 28.1% | 54.5% | 29.3% | 38.0% | 19.5% | |
| Multilingual Q&A MMMLU | 91.8% | 91.8% | 86.6% | 89.5% | 89.1% | 89.6% | 86.8% | |
| Commonsense reasoning across 100 Languages and Cultures Global PIQA | 92.8% | 93.4% | 90.2% | 91.5% | 90.1% | 91.2% | 85.6% | |
| Long context performance MRCR v2 (8-needle) | 128k (average) | 67.2% | 77.0% | 54.3% | 58.0% | 47.1% | 81.9% | 54.6% |
| 1M (pointwise) | 22.1% | 26.3% | 21.0% | 16.4% | not supported | not supported | 6.1% |
Iterative development and design
We’ve seen impressive results on tasks that require building something by making small changes over time.
Aiding scientific and mathematical discovery
By reasoning through complex problems, Deep Think can act as a powerful tool for researchers.
Algorithmic development and code
Deep Think excels at tough coding problems where problem formulation and careful consideration of tradeoffs and time complexity is paramount.
Create and adapt voxel art
Recombine and regenerate voxel art through Gemini 3’s advanced reasoning
Build a procedural fractal world
Create interactive, playable sci-fi worlds through Gemini 3 and Shaders
Vibe code a retro videogame
Code a complex, interactive 3D game, all from a single prompt
Gemini
Supercharge your creativity and productivity
AI Mode
Ask whatever's on your mind to get an AI powered response
Google AI Studio
The fastest path from prompt to production
Google Antigravity
Our new agentic development platform, evolving the IDE into the agent-first era
Gemini API
Get started building with cutting-edge AI models
Vertex AI Studio
Test, tune, and deploy enterprise-ready generative AI