Gemini 3
The best model in the world for multimodal understanding, and our most powerful agentic and vibe-coding model yet, delivering richer visuals and deeper interactivity, all built on a foundation of state-of-the-art reasoning.
Google Multimodal reasoning with large context and agentic tools
Gemini 3 Pro is a multimodal text generation model developed by Google, released in November 2025. It supports a context window of 1,048,576 tokens and is designed to handle complex reasoning tasks, nuanced instruction following, and agentic workflows. The model is available to developers through Google AI Studio and Vertex AI, and is also integrated into Google Search and the Gemini app.
Gemini 3 Pro is built for tasks that require understanding context and intent with minimal prompting, including multi-step problem solving, code generation, and multimodal input processing. It is positioned as Google's primary model for agentic development, including use within the Google Antigravity platform. The model accepts tool inputs alongside text and numeric parameters, making it suited for applications that require dynamic tool use and structured interactions.
What Gemini 3 supports
Large Context Window
Processes up to 1,048,576 tokens in a single request, enabling analysis of long documents, codebases, or extended conversation histories without truncation.
Advanced Reasoning
Applies multi-step reasoning to complex problems, designed to parse layered or ambiguous inputs and infer intent with reduced prompting.
Multimodal Input
Accepts text and image inputs together, allowing the model to interpret visual content alongside written instructions in a single request.
Tool Use
Supports tool-calling inputs natively, enabling integration with external APIs, functions, and agentic workflows through structured tool definitions.
Agentic Task Execution
Designed for multi-step agentic tasks, including autonomous planning and execution sequences used in platforms like Google Antigravity.
Code Generation
Generates, explains, and debugs code across multiple programming languages, with particular emphasis on interactive and vibe-coding use cases.
Ready to build with Gemini 3?
Get Started FreeBenchmark scores
Scores represent accuracy — the percentage of questions answered correctly on each test.
| Benchmark | What it tests | Score |
|---|---|---|
| MMLU-Pro | Expert knowledge across 14 academic disciplines | 89.8% |
| GPQA Diamond | PhD-level science questions (biology, physics, chemistry) | 90.8% |
| LiveCodeBench | Real-world coding tasks from recent competitions | 91.7% |
| HLE | Questions that challenge frontier models across many domains | 37.2% |
| SciCode | Scientific research coding and numerical methods | 56.1% |
| AIME 2025 | American math olympiad problems (2025) | 95.0% |
| ARC-AGI-2 | Novel abstract reasoning and pattern recognition | 31.1% |
| SWE-bench Verified | Real GitHub issues requiring multi-file code fixes | 76.2% |
| MMMLU | Multilingual and multimodal understanding | 91.8% |
Common questions about Gemini 3
What is the context window size for Gemini 3 Pro?
Gemini 3 Pro supports a context window of 1,048,576 tokens, which allows it to process very long documents, extended conversations, or large codebases in a single request.
What is the training data cutoff for Gemini 3 Pro?
Based on the available metadata, Gemini 3 Pro has a training date of November 2025. Specific knowledge cutoff details should be confirmed in the official Google AI documentation.
Where can developers access Gemini 3 Pro?
Gemini 3 Pro is available to developers through Google AI Studio and Vertex AI, as well as through MindStudio without requiring separate API key setup.
What input types does Gemini 3 Pro support?
The model accepts select, number, and tools input types, making it compatible with structured tool-calling workflows in addition to standard text prompts.
Is Gemini 3 Pro suitable for agentic applications?
Yes. Gemini 3 Pro is described by Google as their primary agentic model and is integrated into the Google Antigravity agentic development platform. It supports tool use natively, which is a key requirement for agentic task execution.
What people think about Gemini 3
Community discussion around Gemini 3 Pro on Reddit has been broadly positive, with users highlighting its agentic capabilities and benchmark performance, including a noted result of completing Pokémon Crystal using 50% fewer tokens than its predecessor. The release of the related Gemini 3.1 Pro update generated significant engagement, with one thread accumulating over 2,400 upvotes and 528 comments.
Some threads focus on benchmark comparisons and token efficiency as practical indicators of model capability, while others discuss the Flash variant's cost and arena rankings as points of reference for the broader Gemini 3 family. Overall community interest centers on real-world agentic use cases and cost-performance tradeoffs.
Google releases Gemini 3.1 Pro with Benchmarks
Google just dropped Gemini 3.1 Pro. Mindblowing model.
Google just dropped a new Agentic Benchmark: Gemini 3 Pro beat Pokémon Crystal (defeating Red) using 50% fewer tokens than Gemini 2.5 Pro.
Google releases Gemini 3 Flash: Ranks #3 on LMArena (above Opus 4.5), scores 99.7% on AIME and costs $0.50/1M plus Benchmarks.
Parameters & options
Must be less than Max Response Size
Explore similar models
Google Gemini 3.1 Pro
Google's frontier reasoning model delivering enhanced software engineering, agentic reliability, and multimodal intelligence across a 1M-token context window.
Google Gemini 2.5 Pro
State-of-the-art thinking model, capable of reasoning over complex problems in code, math, and STEM, as well as analyzing large datasets, codebases, and documents using long context.
Google Gemini 3 Flash
A fast, capable thinking model from Google designed for agentic workflows, coding, and multi-turn chat with near Pro-level reasoning at lower latency.
Anthropic Claude 4.6 Sonnet
Anthropic's most capable Sonnet model yet, delivering near-Opus-level intelligence for coding, agents, and computer use with a 1M token context window.
OpenAI GPT‑5.2 Pro
The most capable model series yet for professional knowledge work.
DeepSeek DeepSeek-R1
Reasoning LLM from Chinese AI company DeepSeek utilizing Chain of Thought.
Start building with Gemini 3
No API keys required. Create AI-powered workflows with Gemini 3 in minutes — free.