Benchmark Almanac

Gemini 2.5 Pro Jun 2025

Gemini 2.5 Pro ranks first on 10 of the 192 benchmarks with five or more models scored.

Google DeepMind Released June 5th, 2025 Family Gemini 2.5 The lab's model card Compare with others
211
Benchmarks with a score
10
Ranked first, out of five or more models
34
In the top three
Google DeepMind Jun 2025

Gemini 2.5 Pro

Jun 2025 snapshot Closed weights Flagship

144.9 #96 of 256
Epoch Capabilities Index
Self-Reported Index 143.9 #143 of 404
  • Humanity's Last Exam 21.6% #19 /47
  • SWE-bench Verified 57.6% #30 /32
  • Terminal-Bench 2.0 32.6% #31 /43

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
RewardBench 2 79.5% #2 of 42 Official leaderboard
FaithJudge 6.65% #2 of 37 Official leaderboard
SuperGPQA 66% #4 of 59 Official leaderboard
IndicGenBench 77.1 #1 of 13 Official leaderboard
Fiction.LiveBench 91.7% #5 of 57 Fiction.live leaderboard, via Epoch AI
MMLU-Pro 86% #26 of 212 Official leaderboard
LOFT 87% #1 of 8 Lab model card
Wer wird Millionär? LLM Benchmark 266667.0 #10 of 77 Official leaderboard
FACTS Grounding 87.8% #2 of 15 Lab model card
LLM Confabulation Leaderboard 12.4% #8 of 58 Official leaderboard
SimpleQA 54% #11 of 60 Lab model card
WritingBench 79.3 #9 of 45 Official leaderboard
LMArena (Elo) 1,446 #69 of 335 Official leaderboard
Vectara Hallucination Leaderboard (HHEM) 2.60% #35 of 148 Official leaderboard
EQ-Bench 3 1,523 #19 of 78 Official leaderboard
ECLeKTic 46.8% #2 of 7 Lab model card
FACTS Parametric 63.2% #13 of 45 Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset) 7.00% #32 of 104 Official leaderboard
Global MMLU Lite 93.2% #14 of 43 Official leaderboard
WideSearch 4.30% #2 of 6 Official leaderboard
Creative Writing v3 72.3 #41 of 120 Official leaderboard
MultiNRC 45.1% #14 of 38 Official leaderboard
Humanity's Last Exam 21.6% #19 of 47 Official leaderboard
IFEval 84% #25 of 60 Official leaderboard
Judgemark v2.1 72.1 #23 of 42 Official leaderboard
Longform Creative Writing (EQ-Bench) 58.8 #66 of 119 Official leaderboard
BullshitBench 23.6% #63 of 107 Official leaderboard
MRCR v2 58% #19 of 30 Lab model card
MultiChallenge 53.6% #27 of 42 Official leaderboard
BullshitBench V2 20% #93 of 129 Official leaderboard
FACTS Search 63.9% #23 of 31 Official leaderboard
Phare 60.5% #50 of 62 Official leaderboard
Document Arena 1,433 #35 of 39 Official leaderboard

Math and science

BenchmarkScoreRankSource
PHYBench 45.8% #2 of 28 Official leaderboard
GPQA Main 74.9% #5 of 60 Official leaderboard
HiPhO 70.4% #3 of 36 Official leaderboard
MATH-500 95.2% #5 of 38 Official leaderboard
Misguided Attention 57.4% #9 of 45 Official leaderboard
miniCTX-v2 36.2% #1 of 4 Official leaderboard
AMO-Bench 38.7% #6 of 21 Official leaderboard
SimpleBench 62.4% #34 of 106 Official leaderboard
ForecastBench 0.13 #27 of 82 Official leaderboard
IMO 2025 31.6% #2 of 6 Official leaderboard
Putnam-like 91.2% #2 of 6 Official leaderboard
GPQA Diamond 85.3% #80 of 229 Epoch AI run
ATLAS (OpenCompass) 35.3% #2 of 5 Official leaderboard
OTIS Mock AIME 2024-2025 84.7% #87 of 212 Epoch AI run
AIME 2025 88% #60 of 143 Lab model card
DTBench 82.4% #81 of 180 Official leaderboard
HiddenMath-Hard 80.5% #1 of 2 Lab model card
FrontierMath 14.1% #29 of 55 Official leaderboard
FrontierMath Tier 4 4.17% #29 of 55 Epoch AI run
SciArena 1,010 #18 of 34 Official leaderboard
CORE-Bench 22.2% #11 of 17 Official leaderboard
ARC-AGI-2 4.86% #68 of 98 Official leaderboard
IMO-ProofBench 17.6% #14 of 20 Official leaderboard
General365 48.7% #17 of 23 Official leaderboard
ARC-AGI 41% #71 of 95 Official leaderboard
FrontierMath Tiers 1-3 v2 24.6% #71 of 86 Official leaderboard
MathArena (Apex, IMO, USAMO) 0.52% #41 of 47 Official leaderboard
HMMT Nov 2025 80% #20 of 22 Official leaderboard
FrontierMath Tier 4 v2 0% #67 of 68 Official leaderboard
IMC 2025 94.5% #1 of 1 Official leaderboard

Coding

BenchmarkScoreRankSource
Aider Polyglot 82.2% #2 of 30 Lab model card
LiveCodeBench v6 73.6% #3 of 21 Official leaderboard
ArtifactsBench 57.7% #3 of 20 Official leaderboard
LiveCodeBench 88.9% #4 of 21 Official leaderboard
LiveCodeBench v5 74% #4 of 21 Official leaderboard
AutoCodeBench 48.7% #7 of 34 Official leaderboard
MLE-bench 59.6% #3 of 12 Official leaderboard
LiveSQLBench 28.7% #13 of 36 Official leaderboard
WeirdML 54% #42 of 112 Official leaderboard
METR Time Horizon 40 min #13 of 31 Official leaderboard
IOI (Vals) 17.1% #26 of 55 Official leaderboard
DesignArena Data Visualization 1,220 #56 of 118 Official leaderboard
AlgoTune 1.51× #11 of 18 Official leaderboard
DesignArena Website 1,160 #95 of 151 Official leaderboard
DesignArena Code 1,143 #97 of 143 Official leaderboard
DesignArena UI Component 1,131 #97 of 139 Official leaderboard
DesignArena Game Dev 1,101 #101 of 143 Official leaderboard
CodeClash 1,124 #6 of 8 Official leaderboard
DesignArena 3D Design 1,063 #104 of 136 Official leaderboard
GSO-Bench 3.92% #25 of 30 Official leaderboard
Vibe Code Bench v1 0.40% #15 of 17 Official leaderboard
SWE-Bench Pro Private 10.1% #11 of 12 Official leaderboard
SWE-bench Verified 57.6% #30 of 32 Epoch AI run
SWE-fficiency 3.13% #16 of 17 Official leaderboard
Code Arena 1,228 #121 of 128 Official leaderboard
WebDev Arena 1,226 #100 of 105 arena.ai, via Epoch AI
Vibe Code Bench 0.40% #116 of 120 Official leaderboard
DesignArena Agentic Overall 907 #73 of 74 Official leaderboard
Image-to-WebDev Arena 1,285 #55 of 55 Official leaderboard
FrontierCS 20.3 #14 of 14 Official leaderboard

Agents and tools

BenchmarkScoreRankSource
Step Race 4 #7 of 59 Official leaderboard
Orak 3.90 #1 of 6 Official leaderboard
Meta Agents Research Environments (ARE) / Gaia2 25.8% #3 of 13 Official leaderboard
SnakeBench 29 #45 of 188 Official leaderboard
GAIA 66.1% #23 of 64 Official leaderboard
Terminal-Bench 32.6% #7 of 18 tbench.ai, via Epoch AI
Elimination Game 4 #22 of 56 Official leaderboard
Galileo Agent Leaderboard v2 43% #10 of 21 Official leaderboard
LisanBench 556.0 #66 of 133 Official leaderboard
OSWorld-Verified 45.8% #19 of 33 Official leaderboard
Pencil Puzzle Bench 0% #34 of 53 Official leaderboard
MCP-Universe 22.1% #20 of 30 Official leaderboard
ITBench 42.3% #9 of 13 Official leaderboard
DeepSearchQA 24.6% #7 of 10 Official leaderboard
Agent Poker Bench 1,033 #12 of 17 Official leaderboard
Terminal-Bench 2.0 32.6% #31 of 43 Official leaderboard
PinchBench 63.1% #36 of 48 Official leaderboard
MCPMark 15.8% #21 of 28 Official leaderboard
DeepResearch Bench 41.5% #18 of 22 evals.futuresearch.ai, via Epoch AI
τ³-bench Banking 13.7% #22 of 25 Official leaderboard
CompileBench 60% #18 of 20 Official leaderboard
LMArena Search Arena 1,145 #34 of 37 Official leaderboard
Toolathlon 10.5% #49 of 50 Official leaderboard
LiveResearchBench 61.6 #7 of 7 Official leaderboard
AndroidDaily 36.6% #2 of 2 Official leaderboard

Professional work

BenchmarkScoreRankSource
MedCode 50.6% #17 of 96 Official leaderboard
LegalBench 84.3% #35 of 138 Official leaderboard
CaseLaw v2 63.9% #15 of 53 Official leaderboard
FailureSensorIQ 67% #7 of 24 Official leaderboard
MortgageTax 67.2% #27 of 87 Official leaderboard
ProfBench 52.1% #23 of 72 Official leaderboard
TaxEval v2 72.9% #55 of 129 Official leaderboard
Vals Finance Agent v1 29.4% #25 of 47 Official leaderboard
CorpFin v2 60.8% #64 of 119 Official leaderboard
MedHELM 51.9 #8 of 13 Official leaderboard
PRBench 41.4% #32 of 48 Official leaderboard
Vals Finance Agent 41.6% #39 of 49 Official leaderboard
Vending-Bench 2 $574 #55 of 67 Official leaderboard
Vending-Bench Arena $773 #27 of 32 Official leaderboard
Remote Labor Index 0.83% #19 of 19 Official leaderboard

Vision, audio and robotics

BenchmarkScoreRankSource
VHELM 87.1 #1 of 41 Official leaderboard
VISTA 54.6% #2 of 54 Official leaderboard
ERQA 55.6% #1 of 17 Official leaderboard
AHELM 80.3 #1 of 10 Official leaderboard
Math-VR 64.7% #3 of 26 Official leaderboard
MMMU 82% #12 of 103 Lab model card
ActivityNet-QA 66.7% #1 of 8 Lab model card
SpatialViz-Bench 44.7% #1 of 8 Claim in arxiv.org, via Epoch AI
LVBench 78.7% #5 of 31 Lab model card
Perception Test MCVQA 78.4% #1 of 6 Lab model card
VisualPuzzles 49.5% #5 of 29 Official leaderboard
VATEX 71.3 #1 of 5 Lab model card
MADQA 73.1% #4 of 18 Official leaderboard
GeoBench 81% #12 of 48 Official leaderboard
Blueprint-Bench 0.42 #2 of 8 Official leaderboard
VATEX ZH 59.7 #1 of 4 Lab model card
YouCook2 188.3 #1 of 4 Lab model card
OCR Arena 1,745 #3 of 11 Official leaderboard
Video-MME 86.9% #15 of 51 Lab model card
VPCT 46.4% #9 of 27 VPCT leaderboard, via Epoch AI
Vibe-Eval 67.2% #3 of 9 Lab model card
BetterChartQA 72.4% #1 of 3 Lab model card
Vision Arena 1,247 #47 of 136 Official leaderboard
1H-VideoQA 79% #7 of 20 Official leaderboard
RoboRewardBench 0.90 #8 of 22 Official leaderboard
Audio MC 46.9% #7 of 18 Official leaderboard
Video-MMMU 83.6% #15 of 36 Lab model card
CoVoST2 38.5 #7 of 14 Lab model card
MINERVA 67.6% #3 of 6 Lab model card
Neptune 87.3% #1 of 2 Lab model card
ClockBench 18.9% #20 of 37 Official leaderboard
Math Kangaroo 2025 (MathArena) 77.2% #14 of 22 Official leaderboard
OmniDocBench v1.5 88.0 #7 of 11 Official leaderboard
ZeroBench 5.00% #52 of 76 Official leaderboard
VoxelBench Image 1,199 #10 of 13 Official leaderboard
EgoTempo 44.3% #4 of 5 Lab model card
QVHighlights 75% #4 of 5 Lab model card
MineBench 1,128 #61 of 75 Official leaderboard
Drone-Bench 33.9% #16 of 19 Official leaderboard
VoxelBench 1,282 #53 of 56 Official leaderboard
DECIDE 18.0 #8 of 8 Official leaderboard

Safety and security

BenchmarkScoreRankSource
PCT (Polarized Contrastive Pairs) 37.3% #9 of 46 Official leaderboard
In-house CTF suite (Phuong et al. 2024) 84.6% #1 of 4 Lab model card
Anthropic Red Team 99.5% #20 of 75 Official leaderboard
AIR-Bench 2024 73.6% #31 of 82 Official leaderboard
BaxBench 33.3% #14 of 34 Official leaderboard
WMDP Chemistry 80% #3 of 7 Lab model card
HELM Safety 91.4 #37 of 75 Official leaderboard
LAB-Bench CloningScenarios 53% #6 of 12 Lab model card
WMDP Biology 86% #4 of 8 Lab model card
GDM Stealth 40% #3 of 6 Lab model card
Cyber key skills benchmark v1 50% #2 of 4 Lab model card
SecureBio VMQA, single choice 56% #2 of 4 Lab model card
Dangerous Content policy violations (from ART) 24.3% #1 of 2 Lab model card
Helpfulness violations (from ART) 6.10% #1 of 2 Lab model card
Indirect prompt injection, TAP attack 30.8% #1 of 2 Lab model card
BinaryAudit 21.1% #11 of 20 Official leaderboard
KORA 29.1% #28 of 42 Official leaderboard
LAB-Bench SeqQA 71% #8 of 12 Lab model card
LAB-Bench ProtocolQA 74% #11 of 16 Lab model card
Hack the Box challenge set (Google DeepMind) 7.70% #3 of 4 Lab model card
InterCode-CTF 97.4% #3 of 4 Lab model card
Spiral-Bench v1.2 28.3 #16 of 21 Official leaderboard
SimpleSafetyTests 97% #58 of 75 Official leaderboard
Spiral-Bench v1.0 43.5 #11 of 13 Official leaderboard
FORTRESS 61.7% #71 of 73 Official leaderboard
GDM Situational Awareness 18.2% #8 of 8 Lab model card
Divergence attack (extractable memorization) 59% #2 of 2 Lab model card
Indirect prompt injection, Actor Critic attack 61.4% #2 of 2 Lab model card
Indirect prompt injection, Beam Search attack 63.8% #2 of 2 Lab model card

Other

BenchmarkScoreRankSource
TutorBench 55.6% #3 of 20 Official leaderboard
HELM Capabilities 74.5 #14 of 60 Official leaderboard
Conceptual Reasoning Index 47.5 #79 of 151 Official leaderboard
ACCoRD 62.7 #82 of 152 Official leaderboard
SAGE (Vals) 41.9% #55 of 81 Official leaderboard
CAIS Vision Capabilities Index 45.7 #37 of 53 Official leaderboard
CAIS Text Capabilities Index 22.4 #49 of 68 Official leaderboard
CAIS Risk Index 64.3 #58 of 67 Official leaderboard

Its lab's numbers put it at 143.9, in line with 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 141.4 to 146.4. 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
SWE-bench Verified 67.2% SWE-bench verified, multiple attempts (single attempt = 59.6%) 53.6% Leaderboard, mini-SWE-agent +14 pts
ECLeKTic 46.8% ECLeKTic (Table 3) 36.9% Leaderboard +9.9 pts
Global MMLU Lite 89.2% Global MMLU (Lite) 93.2% Leaderboard −4.0 pts
1H-VideoQA 81% 1H-VideoQA, test, 0-shot (Table 6) 79% Leaderboard +2.0 pts
GPQA Diamond 86.4% GPQA diamond, single attempt (pass@1) 85.3% Epoch AI run +1.1 pts
Aider Polyglot 82.2% Aider Polyglot (diff-fenced) 83.1% Leaderboard, aider, diff-fenced edits −0.9 pts
ZeroBench 4.50% ZeroBench, Gemini as judge, average of 4 runs (Table 3) 5.00% Leaderboard −0.5 pts
Humanity's Last Exam 21.6% Humanity's Last Exam 21.6% Leaderboard Same
FACTS Grounding 87.8% FACTS Grounding 87.8% Leaderboard Same
Video-MMMU 83.6% VideoMMMU 83.6% 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 2.5 Pro model card (updated 2025-06-27) Jun 17, 2025

Its system card

Every result in the longer documents the lab published with it. 46 of them aren't in the launch post.

Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities Jul 7, 2025

Capabilities

Safety

Robustness

Alignment

Bio and chem risk

Cyber risk

AI R&D risk