Gemini 3 Flash

Gemini 3 Flash ranks first on 8 of the 237 benchmarks with five or more models scored.

Google DeepMind Released December 17th, 2025 Family Gemini 3 The lab's model card Compare with others

242

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Google DeepMind Dec 2025

Gemini 3 Flash

Closed weights Mid-size

152.3 #59 of 256

Epoch Capabilities Index

Self-Reported Index 149.5 #100 of 404

SWE-bench Verified 75.4% #13 /32
SWE-Bench Pro 34.6% #15 /24
MMMU-Pro 81.2% #13 /90

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
MMLU-Pro	88.6%	#5 of 212	Official leaderboard
MMMLU	91.8%	#4 of 71	Lab model card
FACTS Search	81%	#3 of 31	Official leaderboard
LMArena (Elo)	1,473	#38 of 335	Official leaderboard
SimpleQA Verified	66.8%	#13 of 94	Epoch AI run
Judgemark v2.1	85.4	#6 of 42	Official leaderboard
Global PIQA	92.8%	#2 of 14	Lab model card
KINA	43.9%	#6 of 37	Official leaderboard
LOCA-Bench	20%	#2 of 9	Official leaderboard
FACTS Parametric	72.3%	#11 of 45	Official leaderboard
Hemingway-bench	1,052	#12 of 42	Official leaderboard
SkateBench v2	76.2%	#6 of 21	Official leaderboard
FACTS Parametric V2	72.4%	#9 of 27	Official leaderboard
Humanity's Last Exam	43.5%	#58 of 154	Lab model card
Extended NYT Connections	83.1%	#39 of 103	Official leaderboard
OpenAI MRCR v2	44.9%	#22 of 47	Official leaderboard
FACTS Multimodal	49.7%	#11 of 23	Official leaderboard
FACTS Search V2	80.9%	#11 of 21	Official leaderboard
MRCR v2	67.2%	#16 of 30	Lab model card
FACTS Benchmark Suite	63.8%	#13 of 23	Official leaderboard
Visual Aesthetic Benchmark	28%	#11 of 18	Official leaderboard
Longform Creative Writing (EQ-Bench)	55.7	#76 of 119	Official leaderboard
Judgemark v4	46.1	#28 of 40	Official leaderboard
Structured Output Benchmark	83.3%	#15 of 19	Official leaderboard
Continual Learning Bench	0.09	#4 of 5	Official leaderboard
Document Arena	1,440	#32 of 39	Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset)	13.5%	#88 of 104	Official leaderboard
BullshitBench V2	10%	#115 of 129	Official leaderboard
BullshitBench	10.9%	#98 of 107	Official leaderboard
FACTS Grounding v2	48.6%	#24 of 25	Official leaderboard

Math and science

Benchmark	Score	Rank	Source
BRUMO 2025	100%	#1 of 42	Official leaderboard
SMT 2025	92.9%	#2 of 40	Official leaderboard
AIME 2025	99.7%	#10 of 143	Lab model card
General365	60.8%	#2 of 23	Official leaderboard
HMMT 2025	97.5%	#5 of 53	Official leaderboard
AIME 2026	96.7%	#6 of 31	Official leaderboard
OTIS Mock AIME 2024-2025	95.6%	#46 of 212	Epoch AI run
FrontierMath	35.6%	#12 of 55	Official leaderboard
GPQA Diamond	89.4%	#54 of 229	Epoch AI run
CMIMC 2025	90%	#9 of 33	Official leaderboard
ForecastBench	0.13	#23 of 82	Official leaderboard
DTBench	89.1%	#56 of 180	Official leaderboard
HMMT Nov 2025	93.3%	#7 of 22	Official leaderboard
MathArena (Apex, IMO, USAMO)	15.6%	#16 of 47	Official leaderboard
SimpleBench	61.1%	#38 of 106	Official leaderboard
SciPredict	22.2%	#5 of 12	Official leaderboard
MathArena Apex Shortlist	68.6%	#17 of 38	Official leaderboard
HMMT 2026	89.4%	#14 of 31	Official leaderboard
ARC-AGI	84.7%	#46 of 95	Official leaderboard
ARC-AGI-2	33.6%	#49 of 98	Official leaderboard
FrontierMath Tiers 1-3 v2	51.2%	#52 of 86	Official leaderboard
FrontierMath Tier 4	4.17%	#34 of 55	Epoch AI run
Human Behavior Bench	78.5	#7 of 11	Official leaderboard
ProofBench	15%	#38 of 59	Official leaderboard
FrontierMath Tier 4 v2	17.1%	#49 of 68	Official leaderboard
Verifying the Verifiers (Project APE CRED benchmark)	47%	#19 of 23	Official leaderboard

Coding

Benchmark	Score	Rank	Source
SWE-bench Multilingual	72.7%	#1 of 13	Official leaderboard
DesignArena Agentic Image to HTML	1,224	#2 of 11	Official leaderboard
IOI (Vals)	39.1%	#13 of 55	Official leaderboard
WeirdML	61.6%	#29 of 112	Official leaderboard
OTelBench	18.8%	#4 of 14	Official leaderboard
ALE-Bench	1,722	#41 of 126	Official leaderboard
CooperBench	27.8%	#2 of 6	Official leaderboard
SWE-fficiency	10.6%	#6 of 17	Official leaderboard
SWE-bench Verified	75.4%	#13 of 32	Epoch AI run
DesignArena Website	1,189	#78 of 151	Official leaderboard
ProgramBench	0%	#13 of 25	Official leaderboard
DesignArena 3D Design	1,166	#71 of 136	Official leaderboard
DesignArena Code	1,180	#78 of 143	Official leaderboard
Code Arena	1,440	#71 of 128	Official leaderboard
GSO-Bench	9.80%	#17 of 30	Official leaderboard
DesignArena Game Dev	1,156	#84 of 143	Official leaderboard
OpenHands Index	49%	#20 of 33	Official leaderboard
SWE-Bench Pro	34.6%	#15 of 24	Official leaderboard
AndroidBench	62.5%	#22 of 32	Official leaderboard
WebDev Arena	1,383	#74 of 105	arena.ai, via Epoch AI
GameDevBench	46.8%	#12 of 17	Official leaderboard
Vibe Code Bench	20.2%	#87 of 120	Official leaderboard
LiveCodeBench Pro	2,316	#3 of 4	Lab model card
DesignArena Agentic Overall	1,074	#56 of 74	Official leaderboard
Image-to-WebDev Arena	1,451	#42 of 55	Official leaderboard
SWE Atlas Test Writing	30.3%	#17 of 22	Official leaderboard
DesignArena Full Stack	1,042	#38 of 49	Official leaderboard
DAXBench v2	63.5%	#39 of 47	Official leaderboard
ts-bench	20%	#6 of 7	Official leaderboard
DesignArena Frontend Web App	1,092	#46 of 53	Official leaderboard
DesignArena React Native	1,084	#49 of 56	Official leaderboard
ProgramBench almost solved	0%	#22 of 25	Official leaderboard
ProgramBench tests passed	31.8%	#22 of 25	Official leaderboard
Code Migration	6.39%	#69 of 78	Official leaderboard
ts-bench v1	92%	#8 of 9	Official leaderboard
DesignArena Android (Kotlin)	1,034	#43 of 46	Official leaderboard
DesignArena Image to Frontend	1,059	#31 of 31	Official leaderboard
SWE Atlas	8.20%	#22 of 22	Official leaderboard
SWE Atlas Refactoring	10%	#17 of 17	Official leaderboard
Vision2Web S1-2026	30.8%	#5 of 5	Official leaderboard

Agents and tools

Benchmark	Score	Rank	Source
PinchBench	87%	#1 of 48	Official leaderboard
VITA-Bench	32.5%	#1 of 18	Official leaderboard
Elimination Game	6	#5 of 56	Official leaderboard
Game Arena Chess Openings	1,263	#1 of 11	Official leaderboard
Game Arena Coin Game	472	#5 of 39	Official leaderboard
SnakeBench	32	#30 of 188	Official leaderboard
Toolathlon	49.4%	#8 of 50	Official leaderboard
Game Arena Dark Hex	359	#7 of 41	Official leaderboard
Agent Poker Bench	1,100	#3 of 17	Official leaderboard
BalatroBench	16.7	#3 of 16	Official leaderboard
tau2-bench	90.2%	#16 of 79	Lab model card
Terminal-Bench 2.0	64.3%	#9 of 43	Official leaderboard
BALROG	48.1%	#7 of 33	Official leaderboard
VAKRA	38.1%	#1 of 4	Official leaderboard
DeepResearch Bench	49.8%	#6 of 22	evals.futuresearch.ai, via Epoch AI
SkillsBench	54.6%	#6 of 22	Official leaderboard
Game Arena Chess	1,213	#12 of 42	Official leaderboard
Deep Research Bench (FutureSearch)	0.50	#6 of 21	Official leaderboard
AdaPlanBench	43.3%	#3 of 10	Official leaderboard
LisanBench	1996.0	#43 of 133	Official leaderboard
GAIA	68.4%	#21 of 64	Official leaderboard
DeepPlanning	33.8%	#5 of 15	Official leaderboard
Game Arena Five in a Row	756	#7 of 20	Official leaderboard
GameWorld	41.9%	#8 of 22	Official leaderboard
MM-BrowseComp	18.5%	#4 of 11	Official leaderboard
Game Arena Werewolf	-1	#3 of 8	Official leaderboard
Pencil Puzzle Bench	6.70%	#21 of 53	Official leaderboard
Game Arena Word Art	677	#16 of 39	Official leaderboard
Chess Puzzles	40%	#39 of 93	Epoch AI run
Mystery Game Puzzles	26%	#38 of 84	Epoch AI run
ScreenSpot-Pro	69.1%	#15 of 32	Lab model card
EnterpriseOps-Gym	34%	#6 of 12	Official leaderboard
CUA-World-Long	14%	#2 of 4	Official leaderboard
Game Arena Word Association	262	#21 of 40	Official leaderboard
PinchBench 2.0	72.1%	#30 of 57	Official leaderboard
RuneBench	4.69	#41 of 76	Official leaderboard
τ³-bench Banking	27.3%	#14 of 25	Official leaderboard
Game Arena Bargaining	180	#22 of 39	Official leaderboard
Clocktower Radio	1,541	#13 of 23	Official leaderboard
Kaggle Game Arena	283	#24 of 42	Official leaderboard
tau3-bench	83.5%	#4 of 7	Official leaderboard
Game Arena Checkers	896	#24 of 41	Official leaderboard
Game Arena Nine Men's Morris	585	#23 of 39	Official leaderboard
OmniGAIA	51.7%	#3 of 5	Official leaderboard
Game Arena Poker	-3.86	#20 of 31	Official leaderboard
Game Arena Reversi	330	#26 of 39	Official leaderboard
iOSWorld	28.6%	#4 of 6	Official leaderboard
Game Arena Dots and Boxes	755	#28 of 40	Official leaderboard
UniClawBench	28.3%	#7 of 10	Official leaderboard
LMArena Search Arena	1,189	#26 of 37	Official leaderboard
OpenComputer	16.4%	#5 of 7	Official leaderboard
Game Arena Four in a Row	258	#31 of 43	Official leaderboard
LLM Chess	-179	#92 of 123	Official leaderboard
Game Arena Go	497	#31 of 39	Official leaderboard
Game Arena Ultimate Tic-Tac-Toe	165	#31 of 39	Official leaderboard
Toloka Arena	22.8%	#4 of 5	Official leaderboard
MCP Atlas	62%	#30 of 37	Official leaderboard
Claw-Eval Pass^3	49.2%	#20 of 24	Official leaderboard
CEO-Bench	$0	#16 of 19	Official leaderboard
ClawProBench	49%	#43 of 49	Official leaderboard
Claw-Eval	72.3%	#22 of 24	Official leaderboard
Game Arena Lines of Action	159	#36 of 39	Official leaderboard
AutoResearchBench	2.75%	#12 of 13	Official leaderboard
Game Arena Clobber	22	#38 of 39	Official leaderboard

Professional work

Benchmark	Score	Rank	Source
MedCode	55.9%	#5 of 96	Official leaderboard
LegalBench	86.9%	#11 of 138	Official leaderboard
MortgageTax	68.7%	#10 of 87	Official leaderboard
APEX-Agents	24%	#1 of 8	Official leaderboard
CorpFin v2	66.4%	#20 of 119	Official leaderboard
TRLawBench	9.46	#8 of 34	Official leaderboard
ProfBench	53.4%	#21 of 72	Official leaderboard
TaxEval v2	73.9%	#44 of 129	Official leaderboard
Vals Finance Agent v1	39.3%	#19 of 47	Official leaderboard
prinzbench	36.0	#13 of 26	Official leaderboard
3DCodeArena	1,010	#7 of 14	Official leaderboard
Vals Index v1	53.2%	#18 of 34	Official leaderboard
CaseLaw v2	55.8%	#32 of 53	Official leaderboard
Vending-Bench Arena	$3,436	#20 of 32	Official leaderboard
PatientAgentBench	3.78	#5 of 8	Official leaderboard
Vals Finance Agent	47.6%	#31 of 49	Official leaderboard
PolicyBench	80.6%	#32 of 50	Official leaderboard
Vending-Bench 2	$3,635	#45 of 67	Official leaderboard
APEX-1	60.9%	#50 of 71	Official leaderboard
Vals Finance Agent v2	42.5%	#60 of 80	Official leaderboard
Legal Research Bench	18.3%	#70 of 93	Official leaderboard
Tax Agent Bench	11.8%	#66 of 84	Official leaderboard
YC-Bench	$393,736	#47 of 59	Official leaderboard
Harvey LAB (Legal Agent Benchmark)	0%	#64 of 80	Official leaderboard
CHI-Bench	12.5%	#19 of 23	Official leaderboard
Vals Index v2	29.4%	#54 of 65	Official leaderboard
MedScribe	69.9%	#85 of 97	Official leaderboard
EnterpriseBench: CoreCraft	20%	#45 of 51	Official leaderboard
Excel Modeling Benchmark	26%	#68 of 76	Official leaderboard
EEBench	19.5%	#28 of 30	Official leaderboard
CommerceAgentBench	29%	#13 of 13	Official leaderboard

Vision, audio and robotics

Benchmark	Score	Rank	Source
MATH-Vision (MathVision)	90.6%	#4 of 105	Official leaderboard
GeoBench	88%	#2 of 48	Official leaderboard
ParseBench	74.1%	#4 of 44	Official leaderboard
OCR Arena	1,781	#1 of 11	Official leaderboard
ChartMuseum	78.6%	#2 of 21	Official leaderboard
VideoMME v2 (with subtitles)	42.5%	#3 of 31	Official leaderboard
Video-MMMU	86.9%	#4 of 36	Lab model card
VPCT	72.6%	#3 of 27	VPCT leaderboard, via Epoch AI
MSEB	39.2	#1 of 9	Official leaderboard
CiteVQA	65.4%	#2 of 16	Official leaderboard
MMMU-Pro	81.2%	#13 of 90	Lab model card
VisFactor	40.5%	#4 of 26	Official leaderboard
OmniDocBench v1.5	90.4	#2 of 11	Official leaderboard
EmoBench-M	65.5%	#3 of 15	Official leaderboard
Vision Arena	1,272	#28 of 136	Official leaderboard
SpatialWorld Digital	38.1%	#2 of 9	Official leaderboard
OmniVideoBench	55.1%	#2 of 8	Official leaderboard
WorldBench	61.8%	#2 of 8	Official leaderboard
MusTBench	38.1%	#1 of 4	Official leaderboard
UNO-Bench	71.4%	#2 of 7	Official leaderboard
1H-VideoQA	79.2%	#6 of 20	Official leaderboard
LVOmniBench	59%	#2 of 6	Official leaderboard
OmniCap-IF	79.5%	#2 of 6	Official leaderboard
FutureOmni	64.8%	#1 of 3	Official leaderboard
ZeroBench	15%	#27 of 76	Official leaderboard
Math Kangaroo 2025 (MathArena)	85.8%	#8 of 22	Official leaderboard
OmniDocBench v1.6	92.6	#3 of 8	Official leaderboard
Claw-Eval-MM	50.4%	#5 of 11	Official leaderboard
CharXiv	80.3%	#36 of 75	Lab model card
RoboRewardBench	0.92	#11 of 22	Official leaderboard
DISBench	12.3%	#5 of 9	Official leaderboard
SpatialWorld Physical	7.20%	#5 of 9	Official leaderboard
vCX Benchmark	83.9%	#27 of 42	Official leaderboard
MineBench	1,244	#53 of 75	Official leaderboard
BVB	62.1	#34 of 45	Official leaderboard
AgenticVBench	17.5%	#7 of 9	Official leaderboard
Blueprint-Bench 2	0%	#35 of 38	Official leaderboard
VoxelBench	1,286	#52 of 56	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
nocot-bench (NCRI)	103.2	#16 of 185	Official leaderboard
BinaryAudit	36.8%	#7 of 20	Official leaderboard
Human Pathogen Capabilities Test v2 (HPCT v2)	56%	#33 of 71	Official leaderboard
CRSBench	56.9%	#3 of 5	Official leaderboard
KORA	24%	#32 of 42	Official leaderboard

Other

Benchmark	Score	Rank	Source
SAGE (Vals)	51.9%	#10 of 81	Official leaderboard
Conceptual Reasoning Index	56.0	#49 of 151	Official leaderboard
LMCA	43.1%	#49 of 147	Official leaderboard
CAIS Vision Capabilities Index	59.3	#18 of 53	Official leaderboard
ACCoRD	68.6	#58 of 152	Official leaderboard
CAIS Text Capabilities Index	36.5	#29 of 68	Official leaderboard
Vals Multimodal Index	52.2%	#24 of 33	Official leaderboard
CAIS Risk Index	66.3	#61 of 67	Official leaderboard

Its lab's numbers put it at 149.5, 2.8 points below its ECI

The Self-Reported Index fits one score to the 17 capability results Google DeepMind reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 147.0 to 152.1. How we compute it.

Its lab's numbers next to 10 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap	What differs
Terminal-Bench 2.0	47.6% Terminal-bench 2.0 (Terminus-2 harness)	64.3% Leaderboard, Junie CLI	−17 pts	Google uses the Terminus-2 harness, while the board pairs Gemini 3 Flash with Junie CLI.
MCP Atlas	57.4% MCP Atlas	62% Leaderboard	−4.6 pts
AIME 2025	99.7% AIME 2025, With code execution; No tools = 95.2%	97.5% Leaderboard, high	+2.2 pts
SWE-bench Verified	78% SWE-bench Verified (Single attempt)	75.8% Leaderboard, high, mini-SWE-agent	+2.2 pts
SimpleQA Verified	68.7% SimpleQA Verified	66.8% Epoch AI run, high	+1.9 pts
FACTS Benchmark Suite	61.9% FACTS Benchmark Suite	63.8% Leaderboard	−1.9 pts
GPQA Diamond	90.4% GPQA Diamond (No tools)	89.4% Epoch AI run, high	+1.0 pts
ARC-AGI-2	33.6% ARC-AGI-2 (ARC Prize Verified)	33.6% Leaderboard, high	Same
Vending-Bench 2	$3,635 Vending-Bench 2 (Net worth (mean))	$3,635 Leaderboard	Same
Toolathlon	49.4% Toolathlon	49.4% Leaderboard	Same

Google DeepMind's models over time

Google DeepMind's models in color, against every other model on the Epoch Capabilities Index, by release date.

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Gemini 3 Flash model card Dec 17, 2025