Google DeepMind has released Gemini 3, their most capable AI model to date, introducing advanced multimodal reasoning and a new Deep Think mode for complex problem-solving.
Key Highlights
- •1 million token context window - Synthesize text, images, video, audio, and code
- •LMArena #1 - 1501 Elo score, topping the leaderboard
- •Deep Think mode - Extended reasoning for complex problems (coming soon to AI Ultra)
Three Core Capabilities
Learn Anything
Advanced multimodal reasoning enables synthesis across all content types with unprecedented context length.
Build Anything
Described as "the best vibe coding and agentic coding model," excelling at zero-shot generation for interactive web UI development.
Plan Anything
Improved long-horizon planning and consistent tool usage for multi-step workflows.
Benchmark Performance
Benchmark | Gemini 3 Pro | Deep Think |
|---|---|---|
| LMArena Elo | 1501 | - |
| Humanity's Last Exam | 37.5% | 41.0% |
| GPQA Diamond | 91.9% | 93.8% |
| MathArena Apex | 23.4% (SOTA) | - |
| MMMU-Pro | 81% | - |
| ARC-AGI-2 | - | 45.1% |
6 items
Availability
Now available:- •Gemini app (all users)
- •AI Mode in Search (Google AI Pro/Ultra)
- •Google AI Studio, Vertex AI, Gemini CLI
- •Google Antigravity (new agentic platform)
- •Gemini 3 Deep Think mode for Google AI Ultra subscribers
Source: Google Blog