Model

Gemini 3 Flash

CLOSED

Google · Gemini 3 family · released Dec 17, 2025

Pro-grade reasoning at Flash latency/cost via an adjustable thinking_level parameter; default model in the Gemini app.

ReasoningCodingVisionFunction callingTool useAgentic

2346.9

Elo · rank #30

Parameters

Undisclosed

Active params

Undisclosed

Context

1M tokens

Architecture

Sparse Mixture-of-Experts Transformer (natively multimodal, based on Gemini 3 Pro)

License

Proprietary (Google API Terms)

Languages

—

API price (in/out)

$0.5 / $3

Modalities

text · vision · audio · video

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath95.2%#16

best: GPT-5.2 · 100.0%

ARC-AGI-2Reasoning33.6%#15

best: GPT-5.6 · 92.5%

CharXivVision80.3%#11

best: Muse Spark 1.1 (xhigh) · 88.4%

FACTS GroundingKnowledge61.9%#11

best: Gemini 2.5 Flash-Lite (Thinking) · 86.8%

GPQA DiamondReasoning90.4%#22

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning33.7%#39

best: Claude Sonnet 5 · 57.4%

LiveCodeBench ProCoding2316#3

best: Gemini 3.1 Pro · 2887

LiveCodeBenchCoding90.8%#4

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MMLUKnowledge91.8%#2

best: OpenAI o3 · 92.9%

MMMU-ProVision81.2%#5

best: Claude Opus 4.7 · 85.5%

SWE-bench VerifiedCoding78.0%#25

best: Claude Fable 5 · 95.0%

Terminal-Bench 2.0Coding47.6%#50

best: GPT-5.6 · 88.8%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$0.5

Output / M tok

Gemini 3 family

Elo progression across releases

Gemini 3 ProNov 20252353.3 Gemini 3 FlashDec 20252346.9

Resources

DOCSai.google.dev/gemini-api/docs/gemini-3 APIAvailable via Google API

API price $0.5/$3 · each benchmark row carries its own source badge (see methodology)