Gemini 3 Flash is Google's high-speed reasoning model that combines near-Pro-level intelligence with the speed and cost efficiency of the Flash line. It outperforms Gemini 2.5 Pro on most benchmarks while running 3× faster and at a fraction of the cost, scoring 78% on SWE-bench Verified. The model supports a 1M-token context window, multimodal inputs (text, images, audio, video, PDFs), configurable thinking levels, and automatic context caching, making it well-suited for agentic workflows, multi-turn chat, and interactive coding assistance.
Google AI FreeGoogle AI PlusGoogle AI ProGoogle AI UltraAPI|VisionReasoningWeb SearchFile|Proprietary Model
Knowledge Cutoff
2025-01-31
Input → Output Format
Context Memory
1.0MIN66KOUT
AI Performance Evaluation
Arena Overall Score
1474
±4As of 2026-05-01
Overall Rank
No.16
30,790 Votes
Arena by Ability
Hard Prompts
1493±6No.16
Expert Knowledge
1496±13No.18
Instruction Following
1458±7No.22
Conversation Memory
1484±9No.18
Creative
1459±10No.11
Coding
1509±8No.26
Math
1476±13No.16
Arena by Occupation
Creative Writing
1462±8No.11
Social Sciences
1485±9No.20
Media
1455±9No.12
Business
1471±8No.17
Healthcare
1485±13No.24
Legal
1488±13No.13
Software
1502±7No.22
Mathematics
1472±16No.21
Source:Arena Intelligence
Overall
AA Intelligence Index
46%↑7%
LiveBench
54%↓6%
ForecastBench
59%↑0%
Reasoning & Math
AA Math Index
97%↑23%
GPQA Diamond
90%↑8%
HLE
35%↑17%
MMLU-Pro
89%↑8%
AIME 2025
97%↑23%
LB Reasoning
49%↓20%
LB Math
68%↓6%
LB Data
48%↓5%
Coding
AA Coding Index
43%↑6%
LiveCodeBench
91%↑25%
LB Coding
79%↑6%
LB Agentic
43%↓2%
TAU2
80%↑0%
TerminalBench
39%↑5%
SciCode
51%↑9%
Language & Instructions
IFBench
78%↑15%
AA-LCR
66%↑4%
Hallucination (HHEM)
14%↑3%
Factual (HHEM)
87%↓3%
LB Language
79%↑6%
LB IF
28%↓23%
Output Speed
Standard Mode
177tok/s↑100
First Output 9.94s
Reasoning Mode
198tok/s↑111
First Output 5.53s
