Robots Atlas>ROBOTS ATLAS
Gemini 3 Flash
AI Models›Gemini

Gemini 3 Flash

3 Flash · Family: Gemini
Gemini 3 Flash is a multimodal language model by Google DeepMind from the Gemini 3 family, designed for fast inference and low cost while retaining reasoning capabilities comparable to Gemini 3 Pro.
⏳ Preview⏳ Limited accessLLMMultimodalReasoning modelTool-using model📁 Gemini
Context window
1M
tokens
Max output
65,536
tokens
Release date
17 December 2025
🔬Google DeepMindResearch lab🏢GoogleOwner
Access:APIHostedDeployment:☁ Cloud

Overview

Gemini 3 Flash is an AI model developed by Google DeepMind, announced on December 17, 2025 as an expansion of the Gemini 3 model family. It is a multimodal model supporting text, image, video, audio, and PDF document inputs, generating text and code as output.

The model has a context window of up to 1 million tokens and a maximum output of 64,000 tokens. It supports tools including function calling, structured output, search as a tool, and code execution. It is available via the Gemini API, Google AI Studio, Vertex AI, Gemini CLI, Android Studio, Google Antigravity, and the Gemini app.

The model's knowledge cutoff is January 2025. It is available in preview. The number of parameters has not been publicly disclosed by the developer.

Classification
LLMMultimodalReasoning modelTool-using model
Family: Gemini
Applications
CodingContent generationDocument analysisWorkflow automationWriting assistanceQ&A / Question answeringResearch assistanceData analysis
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
📏 Context: 1M
✓ Tools
📥 Input: text, image, audio, video…
Platforms
Vertex AI

Technical specification

Context window
1M
tokens
Max output tokens
65,536
tokens per response
Knowledge cutoff
1 Jan 2025
Knowledge boundary
License
proprietary
Hardware requirements
Available only through Google cloud infrastructure (Gemini API, Vertex AI, Google AI Studio).
Features:✓ Tool use
Modalities
⬇ Input
textimageaudiovideodocuments
⬆ Output
textcode

Capabilities and applications

Native model capabilities
Reasoning
Category: reasoning
Multi-step reasoning
Category: reasoning
Long context
Category: reasoning
Multimodal understanding
Category: multimodal
Coding
Category: coding
Function Calling
Category: planning
Structured output
Category: structured_generation
Audio understanding
Category: audio
Image understanding
Category: vision
Video Understanding
Category: video
Chart understanding
Category: vision
Diagram reasoning
Category: reasoning
OCR
Category: vision
Multilingual
Category: language
Planning
Category: planning
Streaming output
Category: reasoning
Interleaved Multimodal Input
Category: reasoning
Application domains
CodingContent generationDocument analysisWorkflow automationWriting assistanceQ&A / Question answeringResearch assistanceData analysisSearch assistance

Benchmark results

15 benchmarks
Humanity's Last Exam
accuracy · No tools, Gemini 3 Flash Thinking
33.7%%
📄 https://deepmind.google/models/gemini/flash/
Full set (text + MM). No tools.
Humanity's Last Exam
accuracy · With search and code execution, Gemini 3 Flash Thinking
43.5%%
📄 https://deepmind.google/models/gemini/flash/
Full set (text + MM). With search and code execution.
GPQA Diamond
accuracy · No tools, Gemini 3 Flash Thinking
90.4%%
📄 https://deepmind.google/models/gemini/flash/
Scientific knowledge, no tools.
ARC-AGI-2
accuracy · ARC Prize Verified, Gemini 3 Flash Thinking
33.6%%
📄 https://deepmind.google/models/gemini/flash/
Abstract reasoning puzzles, ARC Prize verified.
AIME 2025
accuracy · No tools, Gemini 3 Flash Thinking
95.2%%
📄 https://deepmind.google/models/gemini/flash/
Mathematics, no tools.
AIME 2025
accuracy · With code execution, Gemini 3 Flash Thinking
99.7%%
📄 https://deepmind.google/models/gemini/flash/
Mathematics, with code execution.
MMMU-Pro
accuracy · No tools, Gemini 3 Flash Thinking
81.2%%
📄 https://deepmind.google/models/gemini/flash/
Multimodal understanding and reasoning, no tools.
SWE-Bench Verified
accuracy · Single attempt, Gemini 3 Flash Thinking
78.0%%
📄 https://deepmind.google/models/gemini/flash/
Agentic coding, single attempt.
MMMLU
accuracy · Gemini 3 Flash Thinking
91.8%%
📄 https://deepmind.google/models/gemini/flash/
Multilingual Q&A.
Video-MMMU
accuracy · Gemini 3 Flash Thinking
86.9%%
📄 https://deepmind.google/models/gemini/flash/
Knowledge acquisition from videos.
FACTS Benchmark Suite
accuracy · Gemini 3 Flash Thinking
61.9%%
📄 https://deepmind.google/models/gemini/flash/
Factuality benchmark across grounding, parametric knowledge, search, and MM.
SimpleQA Verified
accuracy · Gemini 3 Flash Thinking
68.7%%
📄 https://deepmind.google/models/gemini/flash/
Parametric knowledge.
τ2-bench
accuracy · Gemini 3 Flash Thinking
90.2%%
📄 https://deepmind.google/models/gemini/flash/
Agentic tool use.
Toolathlon
accuracy · Gemini 3 Flash Thinking
49.4%%
📄 https://deepmind.google/models/gemini/flash/
Long horizon real-world software tasks.
MCP Atlas
accuracy · Gemini 3 Flash Thinking
57.4%%
📄 https://deepmind.google/models/gemini/flash/
Multi-step workflows using MCP.

Pricing

Technical architecture

Deployment and security

☁ Available on platforms
☁Vertex AIPlatform
🔒 Security / Enterprise
✓ Verified enterprise information

Gemini 3 Flash dostępny w Vertex AI i Gemini Enterprise. Model przeszedł ewaluacje bezpieczeństwa zgodne z Frontier Safety Framework Google DeepMind. Model card dostępny publicznie.

Updated: 1 May 2026↗ Security documentation

Sources and related pages