Benchmark Almanac

Gemini 3 Pro

Gemini 3 Pro ranks first on 39 of the 262 benchmarks with five or more models scored.

Google DeepMind Released November 18th, 2025 Family Gemini 3 The lab's model card Compare with others
275
Benchmarks with a score
39
Ranked first, out of five or more models
74
In the top three
Google DeepMind Nov 2025

Gemini 3 Pro

Closed weights Flagship

153.0 #55 of 256
Epoch Capabilities Index
Self-Reported Index 150.1 #92 of 404
  • Humanity's Last Exam 37.5% #8 /47
  • SWE-bench Verified 72.9% #21 /32
  • SWE-Bench Pro 43.3% #8 /24

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

BenchmarkScoreRankSource
MMLU-Pro 90.1% #2 of 212 Official leaderboard
SuperGPQA 73.8% #1 of 59 Official leaderboard
MMMLU 91.8% #3 of 71 Lab model card
LMArena (Elo) 1,486 #18 of 335 Official leaderboard
HUMAINE 34 #3 of 50 Official leaderboard
Judgemark v2.1 86.6 #3 of 42 Official leaderboard
Global PIQA 93.4% #1 of 14 Lab model card
Extended NYT Connections 94.4% #9 of 103 Official leaderboard
AMemGym 55.6% #1 of 11 Official leaderboard
EQ-Bench 3 1,553 #9 of 78 Official leaderboard
SimpleQA Verified 72.1% #7 of 56 Official leaderboard
MultiNRC 59% #6 of 38 Official leaderboard
Visual Aesthetic Benchmark 35% #3 of 18 Official leaderboard
Humanity's Last Exam 37.5% #8 of 47 Official leaderboard
Global MMLU Lite 94.5% #8 of 43 Official leaderboard
Phare 73.3% #12 of 62 Official leaderboard
IndicGenBench 59.9 #3 of 13 Official leaderboard
IFEval 87.7% #14 of 60 Official leaderboard
Hemingway-bench 1,056 #10 of 42 Official leaderboard
SkateBench v2 76.2% #5 of 21 Official leaderboard
FACTS Benchmark Suite 70.5% #1 of 4 Lab model card
MultiChallenge 65.7% #11 of 42 Official leaderboard
WritingBench 78.5 #12 of 45 Official leaderboard
Creative Writing v3 74.3 #34 of 120 Official leaderboard
AraGen 64.2% #15 of 42 Official leaderboard
Longform Creative Writing (EQ-Bench) 68.8 #44 of 119 Official leaderboard
MRCR v2 77% #12 of 30 Lab model card
BullshitBench V2 48% #54 of 129 Official leaderboard
BullshitBench 30.9% #48 of 107 Official leaderboard
Role-play Benchmark (MiniMax) 75.6 #4 of 8 Official leaderboard
Document Arena 1,451 #27 of 39 Official leaderboard
CL-bench 15.8% #18 of 24 Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset) 13.6% #89 of 104 Official leaderboard
HealthBench-Hallu 7.10% #3 of 3 Official leaderboard

Math and science

BenchmarkScoreRankSource
GPQA Main 80.3% #1 of 60 Official leaderboard
SMT 2025 93.4% #1 of 40 Official leaderboard
MATH-500 96.4% #1 of 38 Official leaderboard
HiPhO 78.3% #1 of 36 Official leaderboard
AIME 2025 100% #5 of 143 Lab model card
PHYBench 50.4% #1 of 28 Official leaderboard
General365 62.8% #1 of 23 Official leaderboard
HMMT 2025 97.5% #3 of 53 Official leaderboard
SGI-Bench 33.8% #1 of 15 Official leaderboard
SciPredict 25.3% #1 of 12 Official leaderboard
SciArena 1,086 #4 of 34 Official leaderboard
BRUMO 2025 98.3% #5 of 42 Official leaderboard
GPQA Diamond 92.6% #29 of 229 Epoch AI run
SimpleBench 76.4% #16 of 106 Official leaderboard
FrontierMath 37.6% #10 of 55 Official leaderboard
HMMT Nov 2025 93.3% #4 of 22 Official leaderboard
ForecastBench 0.12 #16 of 82 Official leaderboard
FrontierMath Tier 4 18.8% #11 of 55 Epoch AI run
CMIMC 2025 90% #8 of 33 Official leaderboard
FrontierScience-Olympiad 76.1% #2 of 7 Official leaderboard
CORE-Bench 40% #5 of 17 Official leaderboard
MathArena (Apex, IMO, USAMO) 23.4% #14 of 47 Official leaderboard
IMO-ProofBench 30% #6 of 20 Official leaderboard
OTIS Mock AIME 2024-2025 91.4% #64 of 212 Epoch AI run
xbench-ScienceQA 71.6% #12 of 29 Official leaderboard
ProofBench 20% #30 of 59 Official leaderboard
ARC-AGI-2 31.1% #50 of 98 Official leaderboard
MathArena Apex Shortlist 66.5% #20 of 38 Official leaderboard
ARC-AGI 75% #51 of 95 Official leaderboard
LABBench2 56.4% #3 of 5 Official leaderboard
HMMT 2026 86.4% #20 of 31 Official leaderboard
Leipzig Benchmark 14% #13 of 19 Official leaderboard
FrontierScience-Research 12.4% #5 of 7 Official leaderboard
Putnam 2025 91.0 #3 of 4 Official leaderboard
AIME 2026 91.7% #25 of 31 Official leaderboard

Coding

BenchmarkScoreRankSource
Spider 2.0 94.2% #1 of 23 Official leaderboard
MLE-bench 64.4% #1 of 12 Official leaderboard
OJBench 70% #1 of 8 Official leaderboard
AInsteinBench 42.8% #1 of 6 Official leaderboard
WeirdML 69.9% #21 of 112 Official leaderboard
AlgoTune 1.83× #4 of 18 Official leaderboard
ts-bench v1 96% #2 of 9 Official leaderboard
CF-Div2-Stepfun 83.5% #2 of 8 Official leaderboard
DesignArena SVG 1,235 #26 of 103 Official leaderboard
IOI (Vals) 38.8% #14 of 55 Official leaderboard
DesignArena Agentic Image to HTML 1,222 #3 of 11 Official leaderboard
DesignArena UI Component 1,271 #39 of 139 Official leaderboard
METR Time Horizon 3.7 h #8 of 26 Collected by Epoch AI, source not named
METR Time Horizon 1.1 3.7 h #8 of 26 Official leaderboard
ALE-Bench 1,729 #40 of 126 Official leaderboard
SWE-Bench Pro 43.3% #8 of 24 Official leaderboard
OTelBench 15.9% #5 of 14 Official leaderboard
Snorkel Agentic Coding 51.6% #4 of 11 Official leaderboard
DesignArena Website 1,246 #58 of 151 Official leaderboard
SWE-bench Multilingual 68.7% #5 of 13 Official leaderboard
DesignArena Data Visualization 1,236 #47 of 118 Official leaderboard
DesignArena Code 1,234 #58 of 143 Official leaderboard
DesignArena 3D Design 1,206 #56 of 136 Official leaderboard
SWE-fficiency 10.2% #7 of 17 Official leaderboard
GSO-Bench 18.6% #14 of 30 Official leaderboard
PostTrainBench 18.1% #11 of 23 Official leaderboard
LiveCodeBench Pro 2,439 #2 of 4 Lab model card
DesignArena Game Dev 1,186 #72 of 143 Official leaderboard
Vibe Code Bench v1 14.3% #9 of 17 Official leaderboard
WebDev Arena 1,439 #56 of 105 arena.ai, via Epoch AI
Code Arena 1,441 #70 of 128 Official leaderboard
FeatureBench 45.1% #9 of 16 Official leaderboard
FeatureBench Lite 45.1% #9 of 16 Official leaderboard
FrontierCS 33.1 #8 of 14 Official leaderboard
OpenHands Index 49% #19 of 33 Official leaderboard
GameDevBench 53.8% #10 of 17 Official leaderboard
SWE-bench Verified 72.9% #21 of 32 Epoch AI run
CooperBench 21.8% #4 of 6 Official leaderboard
OctoCodingbench 22.9% #4 of 6 Official leaderboard
DesignArena Full Stack 1,053 #36 of 49 Official leaderboard
SWE-Bench Pro Private 17.9% #9 of 12 Official leaderboard
DesignArena Agentic Overall 1,073 #57 of 74 Official leaderboard
Vibe Code Bench 14.3% #96 of 120 Official leaderboard
VIBE (MiniMax) 82.4% #4 of 5 Official leaderboard
Vision2Web S1-2026 36.1% #4 of 5 Official leaderboard
Image-to-WebDev Arena 1,441 #46 of 55 Official leaderboard
DesignArena Frontend Web App 1,096 #45 of 53 Official leaderboard
APEX-SWE 26.8% #57 of 65 Official leaderboard
Data Agent Benchmark 46.6% #16 of 18 Official leaderboard
DesignArena Image to Frontend 1,104 #29 of 31 Official leaderboard
DesignArena React Native 1,070 #53 of 56 Official leaderboard
DesignArena Android (Kotlin) 1,020 #45 of 46 Official leaderboard

Agents and tools

BenchmarkScoreRankSource
SnakeBench 38 #2 of 188 Official leaderboard
MCP-Universe 44.6% #1 of 30 Official leaderboard
BFCL V4 72.5% #3 of 65 Official leaderboard
BalatroBench 19.3 #1 of 16 Official leaderboard
Step Race 5 #4 of 59 Official leaderboard
MCPMark 53.9% #2 of 28 Official leaderboard
ITBench 58.3% #1 of 13 Official leaderboard
VITA-Bench 31.5% #2 of 18 Official leaderboard
BALROG 58.1% #4 of 33 Official leaderboard
Terminal-Bench 2.0 69.4% #6 of 43 Official leaderboard
MobileWorld 47.4% #3 of 18 Official leaderboard
GAIA 78.1% #11 of 64 Official leaderboard
Game Arena Chess Openings 1,253 #2 of 11 Official leaderboard
LisanBench 4677.0 #26 of 133 Official leaderboard
OmniGAIA 62.5% #1 of 5 Official leaderboard
LLM Chess 1,034 #25 of 123 Official leaderboard
Pencil Puzzle Bench 16.7% #14 of 53 Official leaderboard
MM-BrowseComp 18.8% #3 of 11 Official leaderboard
Elimination Game 5 #16 of 56 Official leaderboard
SnorkelWordle 91% #6 of 19 Official leaderboard
tau2-bench 85.4% #25 of 79 Lab model card
AgencyBench 46.9% #3 of 9 Official leaderboard
Agent Poker Bench 1,079 #6 of 17 Official leaderboard
ScreenSpot-Pro 72.7% #12 of 32 Lab model card
tau2-bench Verified 79.4% #5 of 11 Official leaderboard
LMArena Search Arena 1,201 #17 of 37 Official leaderboard
X Browse (xAI) 26.5% #2 of 4 Official leaderboard
DeepPlanning 27.4% #8 of 15 Official leaderboard
Toolathlon 36.4% #27 of 50 Official leaderboard
RWSearch 74.5% #4 of 7 Official leaderboard
WorldTravel 14.7% #4 of 7 Official leaderboard
DeR2 Bench 80.9% #3 of 5 Official leaderboard
DeepResearch Bench 46.3% #14 of 22 evals.futuresearch.ai, via Epoch AI
MCP Atlas 70.3% #24 of 37 Official leaderboard
Deep Research Bench (FutureSearch) 0.46 #14 of 21 Official leaderboard
EnterpriseOps-Gym 29.7% #8 of 12 Official leaderboard
tau3-bench 82.5% #5 of 7 Official leaderboard
RuneBench 3.80 #58 of 76 Official leaderboard
Terminal-Bench 2.1 73.9% #10 of 13 Official leaderboard
TextQuests 21.4 #22 of 28 Official leaderboard
τ³-bench Banking 18% #20 of 25 Official leaderboard

Professional work

BenchmarkScoreRankSource
LegalBench 87% #7 of 138 Official leaderboard
MortgageTax 69.1% #7 of 87 Official leaderboard
FailureSensorIQ 69.1% #2 of 24 Official leaderboard
FinQA (Snorkel OpenEnv environment) 80.3% #3 of 22 Official leaderboard
SnorkelFinance 80.3% #3 of 22 Official leaderboard
MedCode 52.2% #14 of 96 Official leaderboard
PLawBench 66.3% #2 of 11 Official leaderboard
Vending-Bench $4,388 #2 of 9 Official leaderboard
Vals Finance Agent v1 44.1% #11 of 47 Official leaderboard
LEXam 55.4% #8 of 30 Official leaderboard
Vals Finance Agent 55.1% #14 of 49 Official leaderboard
Finance Reasoning (Snorkel) 46.8% #7 of 21 Official leaderboard
ProfBench 51.7% #25 of 72 Official leaderboard
CorpFin v2 63.7% #43 of 119 Official leaderboard
APEX-Agents 18.4% #3 of 8 Official leaderboard
Vending-Bench 2 $5,478 #30 of 67 Official leaderboard
TaxEval v2 72.6% #60 of 129 Official leaderboard
TaxCalcBench 36.3% #8 of 17 Official leaderboard
prinzbench 35.0 #14 of 26 Official leaderboard
Vals Index v1 52.1% #20 of 34 Official leaderboard
FoodTruck Bench $17,199 #35 of 59 Official leaderboard
Vending-Bench Arena $3,384 #21 of 32 Official leaderboard
Finch (FinWorkBench) 27.3% #4 of 6 Official leaderboard
APEX-1 60.9% #49 of 71 Official leaderboard
PRBench 40.6% #36 of 48 Official leaderboard
CaseLaw v2 53% #42 of 53 Official leaderboard
GDPval-MM 28.1% #5 of 6 Official leaderboard
MedScribe 72% #82 of 97 Official leaderboard
EnterpriseBench: CoreCraft 14.4% #47 of 51 Official leaderboard
Remote Labor Index 1.25% #18 of 19 Official leaderboard
DesignArena Python-PPTX Slides 1,003 #35 of 36 Official leaderboard
OfficeQA Pro 3.00% #12 of 12 Official leaderboard

Vision, audio and robotics

BenchmarkScoreRankSource
VideoMME v2 (with subtitles) 49.4% #1 of 31 Official leaderboard
MMSI-Bench 49.2% #1 of 28 Official leaderboard
VPCT 91% #1 of 27 VPCT leaderboard, via Epoch AI
GSM8K-V 59.1% #1 of 24 Official leaderboard
WorldVQA 47.4% #1 of 22 Official leaderboard
MADQA 85.2% #1 of 18 Official leaderboard
EmoBench-M 70.5% #1 of 15 Official leaderboard
MATH-Vision (MathVision) 86.6% #8 of 105 Official leaderboard
Video-MMMU 87.6% #3 of 36 Lab model card
MDPBench 86.4 #1 of 12 Official leaderboard
MotionBench 70.4% #1 of 12 Official leaderboard
Vision Arena 1,289 #12 of 136 Official leaderboard
CaP-X (CaP-Bench) 32.3% #1 of 10 Official leaderboard
VisualPuzzles 52.7% #3 of 29 Official leaderboard
VisFactor 45.8% #3 of 26 Official leaderboard
OmniVideoBench 61.8% #1 of 8 Official leaderboard
VISTA 51.5% #7 of 54 Official leaderboard
UNO-Bench 76.5% #1 of 7 Official leaderboard
MMMU-Pro 81% #15 of 90 Lab model card
GeoBench 84% #8 of 48 Official leaderboard
LVOmniBench 65.8% #1 of 6 Official leaderboard
VideoDR (with search) 76% #1 of 6 Official leaderboard
RoboRewardBench 0.85 #4 of 22 Official leaderboard
AV-SpeakerBench 77.6% #1 of 5 Official leaderboard
Video Reality Test 76.3% #1 of 5 Official leaderboard
DISBench 24.6% #2 of 9 Official leaderboard
MMLongBench-Doc 60.5% #4 of 17 Official leaderboard
OmniDocBench v1.6 92.9 #2 of 8 Official leaderboard
AVMeme Exam 81.9% #1 of 4 Official leaderboard
OmniDocBench v1.5 90.2 #3 of 11 Official leaderboard
ZeroBench 20% #21 of 76 Official leaderboard
Audio MC 54.6% #5 of 18 Official leaderboard
CharXiv 81.4% #27 of 75 Lab model card
Math Kangaroo 2025 (MathArena) 84.2% #9 of 22 Official leaderboard
ClockBench 31.1% #17 of 37 Official leaderboard
VoxelBench Image 1,411 #6 of 13 Official leaderboard
BabyVision 49.7% #13 of 27 Official leaderboard
OCRBench v2 63.4% #7 of 14 Official leaderboard
MusTBench 36.6% #2 of 4 Official leaderboard
MineBench 1,375 #46 of 75 Official leaderboard
VoxelBench 1,355 #49 of 56 Official leaderboard
OCR Arena 1,367 #10 of 11 Official leaderboard

Safety and security

BenchmarkScoreRankSource
ABC Bench Fragment Design 100% #1 of 27 Official leaderboard
LAB-Bench CloningScenarios 94.2% #1 of 12 Lab model card
BinaryAudit 43.9% #4 of 20 Official leaderboard
LABBench2 ProtocolQA2 61.6% #1 of 5 Official leaderboard
LABBench2 SeqQA2 52.5% #1 of 5 Official leaderboard
Cyber key skills benchmark v1 91.7% #1 of 4 Lab model card
CyberGym-E2E (615 tasks) 22.6% #1 of 4 Official leaderboard
SecureBio VMQA, single choice 61.2% #1 of 4 Lab model card
WMDP Chemistry 84.8% #2 of 7 Lab model card
LAB-Bench ProtocolQA 87.4% #5 of 16 Lab model card
Agentic BAIM-LLM Evaluation (ABLE) 74.7% #6 of 18 Official leaderboard
Lakera B3 (Backbone Breaker Benchmark) 67.1% #12 of 35 Official leaderboard
AutoBaxBench 34.5% #4 of 11 Official leaderboard
WMDP Biology 88.2% #3 of 8 Lab model card
AIR-Bench 2024 73.2% #32 of 82 Official leaderboard
LAB-Bench SeqQA 87.6% #5 of 12 Lab model card
HELM Safety 92.5 #35 of 75 Official leaderboard
Spiral-Bench v1.2 38.7 #10 of 21 Official leaderboard
SusVibes 12.9% #6 of 10 Official leaderboard
KORA 29.5% #27 of 42 Official leaderboard
ABC Bench Liquid-Handling Robot 70% #24 of 34 Official leaderboard
FORTRESS 41.7% #52 of 73 Official leaderboard
SimpleSafetyTests 97.5% #54 of 75 Official leaderboard
DTap Direct Misuse 47.5% #6 of 8 Official leaderboard
GDM Situational Awareness 27.3% #6 of 8 Lab model card
Anthropic Red Team 97% #59 of 75 Official leaderboard
LABBench2 LitQA3 74.4% #4 of 5 Official leaderboard
GDM Stealth 25% #5 of 6 Lab model card
DTap Prompt Injection 54% #8 of 8 Official leaderboard
LABBench2 FigQA2 23.8% #5 of 5 Official leaderboard
Chain-of-thought comprehensibility 99.98% #1 of 1 Lab model card
Chain-of-thought informativeness 99.8% #1 of 1 Lab model card
Cyber key skills benchmark v2 0% #1 of 1 Lab model card

Other

BenchmarkScoreRankSource
HELM Capabilities 79.9 #5 of 60 Official leaderboard
CAIS Text Capabilities Index 39.7 #21 of 68 Official leaderboard
CAIS Vision Capabilities Index 57.1 #20 of 53 Official leaderboard
SAGE (Vals) 47.6% #31 of 81 Official leaderboard
TutorBench 53.7% #9 of 20 Official leaderboard
CAIS Risk Index 64.0 #57 of 67 Official leaderboard

Its lab's numbers put it at 150.1, 2.9 points below its ECI

The Self-Reported Index fits one score to the 16 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 147.2 to 153.1. How we compute it.

Its lab's numbers next to 11 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 54.2% Terminal-Bench 2.0 (Terminus-2 agent) 69.4% Leaderboard, Ante −15 pts The lab used Terminus-2, while the board reports the Ante agent.
Humanity's Last Exam 45.8% Humanity's Last Exam, With search and code execution; No tools = 37.5% 37.5% Leaderboard +8.3 pts
AIME 2025 100% AIME 2025, With code execution; No tools = 95.0% 95% Leaderboard +5.0 pts
SWE-bench Verified 76.2% SWE-Bench Verified (Single attempt) 77.4% Leaderboard, live-SWE-agent −1.2 pts
GPQA Diamond 91.9% GPQA Diamond (No tools) 92.6% Epoch AI run −0.7 pts
MathArena (Apex, IMO, USAMO) 23.4% MathArena Apex 23.4% Leaderboard Same
tau2-bench 85.4% τ2-bench 85.4% Leaderboard Same
ARC-AGI-2 31.1% ARC-AGI-2 (ARC Prize Verified) 31.1% Leaderboard Same
MMMU-Pro 81% MMMU-Pro 81% Leaderboard Same
SimpleQA Verified 72.1% SimpleQA Verified 72.1% Leaderboard Same
Vending-Bench 2 $5,478 Vending-Bench 2 (Net worth (mean)) $5,478 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.

50 100 150 2023 2024 2025 2026 2027 Gemini 4 Carbon Gemini 3.8 Flash Gemini 3.5 Flash-Lite Gemini 3 Pro Gemini 2.5 Flash Gemini 2.5 Flash-Lite Gemini 1.5 Pro 50 100 150 2023 2024 2025 2026 2027 Gemini 4 Carbon Gemini 3.8 Flash

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 Pro model card Nov 18, 2025

Its system card

Every result in the longer documents the lab published with it. 17 of them aren't in the launch post.

Frontier Safety Framework Report - Gemini 3 Pro Nov 18, 2025

Alignment

Monitorability

Bio and chem risk

Cyber risk

AI R&D risk

Persuasion