Benchmark Almanac

GPT-5.2

GPT-5.2 ranks first on 20 of the 272 benchmarks with five or more models scored.

OpenAI Released December 11th, 2025 Family GPT-5 The lab's model card Compare with others
281
Benchmarks with a score
20
Ranked first, out of five or more models
47
In the top three
OpenAI Dec 2025

GPT-5.2

Closed weights Flagship

154.5 #50 of 256
Epoch Capabilities Index
Self-Reported Index 153.2 #74 of 404
  • Humanity's Last Exam 27.8% #14 /47
  • SWE-bench Verified 73.8% #17 /32
  • SWE-Bench Pro 29.9% #16 /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 87.4% #11 of 212 Official leaderboard
Wer wird Millionär? LLM Benchmark 333333.0 #7 of 77 Official leaderboard
LOCA-Bench 21.3% #1 of 9 Official leaderboard
SuperGPQA 61.3% #9 of 59 Official leaderboard
EQ-Bench 3 1,544 #12 of 78 Official leaderboard
AMemGym 50.3% #2 of 11 Official leaderboard
Judgemark v2.1 82.0 #10 of 42 Official leaderboard
LMArena (Elo) 1,437 #81 of 335 Official leaderboard
Phare 71% #15 of 62 Official leaderboard
KINA 39.5% #9 of 37 Official leaderboard
Creative Writing v3 74.0 #35 of 120 Official leaderboard
Humanity's Last Exam 27.8% #14 of 47 Official leaderboard
Longform Creative Writing (EQ-Bench) 71.6 #37 of 119 Official leaderboard
HLE-Verified (text-only Full) 43.3% #2 of 6 Official leaderboard
HealthBench-Hallu 3.80% #1 of 3 Official leaderboard
Extended NYT Connections 83.6% #37 of 103 Official leaderboard
MultiNRC 42.2% #15 of 38 Official leaderboard
Opening Lines Arena 1,559 #6 of 15 Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset) 8.40% #43 of 104 Official leaderboard
FACTS Search 72.2% #13 of 31 Official leaderboard
SkateBench v2 69% #9 of 21 Official leaderboard
Lech Mazur Creative Writing 0.29 #23 of 50 Official leaderboard
BullshitBench 27.3% #51 of 107 Official leaderboard
CL-bench 18.2% #12 of 24 Official leaderboard
Hallucinations on production-representative prompts 3.10% #1 of 2 Lab model card
BullshitBench V2 38% #67 of 129 Official leaderboard
OpenAI MRCR v2 77% #8 of 14 Lab model card
DesignArena ASCII Art 1,147 #47 of 77 Official leaderboard
SimpleQA Verified 37.1% #58 of 94 Epoch AI run
Corp Speak Arena 1,457 #16 of 22 Official leaderboard
Visual Aesthetic Benchmark 24% #15 of 18 Official leaderboard
Hemingway-bench 954 #37 of 42 Official leaderboard
Joke Arena 1,454 #23 of 25 Official leaderboard
Document Arena 1,425 #36 of 39 Official leaderboard

Math and science

BenchmarkScoreRankSource
HMMT 2025 100% #1 of 53 Official leaderboard
AIME 2025 100% #6 of 143 Lab model card
HMMT Nov 2025 99.2% #1 of 22 Official leaderboard
HiPhO 74.4% #2 of 36 Official leaderboard
CMIMC 2025 95.6% #2 of 33 Official leaderboard
FrontierMath 40.7% #5 of 55 Official leaderboard
HMMT 2026 97% #3 of 31 Official leaderboard
SMT 2025 92% #4 of 40 Official leaderboard
AIME 2026 98.3% #4 of 31 Official leaderboard
BRUMO 2025 98.3% #6 of 42 Official leaderboard
FrontierScience-Olympiad 77.1% #1 of 7 Official leaderboard
FrontierScience-Research 25.2% #1 of 7 Official leaderboard
GPQA Diamond 91.4% #34 of 229 Epoch AI run
FrontierMath Tier 4 18.8% #10 of 55 Epoch AI run
OTIS Mock AIME 2024-2025 96.1% #42 of 212 Epoch AI run
LABBench2 60.9% #1 of 5 Official leaderboard
DTBench 90.9% #41 of 180 Official leaderboard
IMO-ProofBench 35.7% #5 of 20 Official leaderboard
MathArena Apex Shortlist 79.3% #10 of 38 Official leaderboard
FrontierMath Tiers 1-3 v2 67.4% #32 of 86 Official leaderboard
MathArena (Apex, IMO, USAMO) 13.5% #18 of 47 Official leaderboard
ARC-AGI-2 52.9% #43 of 98 Official leaderboard
IMProofBench 32.4% #4 of 9 Official leaderboard
ForecastBench 0.13 #38 of 82 Official leaderboard
Echo General AI Prediction Leaderboard 1,001 #8 of 17 Official leaderboard
ARC-AGI 86.2% #45 of 95 Official leaderboard
FrontierMath Tier 4 v2 31.7% #33 of 68 Official leaderboard
SciPredict 20.6% #6 of 12 Official leaderboard
MaCBench 72.6% #1 of 2 Official leaderboard
SpatialBench (LatchBio) 50.1% #8 of 14 Official leaderboard
SimpleBench 45.8% #66 of 106 Official leaderboard
ProofBench 15% #37 of 59 Official leaderboard
LiveMathematicianBench 24.6% #12 of 17 Official leaderboard

Coding

BenchmarkScoreRankSource
AlgoTune 2.05× #1 of 18 Official leaderboard
Vibe Code Bench v1 41.3% #1 of 17 Official leaderboard
OpenAI PRs 55% #1 of 12 Lab model card
OTelBench 26.1% #2 of 14 Official leaderboard
IOI (Vals) 54.8% #8 of 55 Official leaderboard
METR Time Horizon 5.9 h #4 of 26 Collected by Epoch AI, source not named
METR Time Horizon 1.1 5.9 h #4 of 26 Official leaderboard
WeirdML 72.2% #18 of 112 Official leaderboard
DataPrep-Bench Law 76.4 #1 of 5 Official leaderboard
DataPrep-Bench Math 25.5 #1 of 5 Official leaderboard
SWE-fficiency 14.8% #4 of 17 Official leaderboard
PRDBench 69% #1 of 4 Official leaderboard
OpenHands Index 58.8% #10 of 33 Official leaderboard
ALE-Bench 1,715 #42 of 126 Official leaderboard
GSO-Bench 27.4% #11 of 30 Official leaderboard
PostTrainBench 21.4% #9 of 23 Official leaderboard
DataPrep-Bench General 77.9 #2 of 5 Official leaderboard
DataPrep-Bench Medical 39.6 #2 of 5 Official leaderboard
DataPrep-Bench Science 24.8 #2 of 5 Official leaderboard
WebDev Arena 1,480 #47 of 105 WebDev Arena Leaderboard, via Epoch AI
Snorkel Agentic Coding 49.4% #5 of 11 Official leaderboard
MLE-bench 30 16% #4 of 8 Lab model card
DesignArena Game Dev 1,185 #74 of 143 Official leaderboard
DesignArena Website 1,188 #79 of 151 Official leaderboard
Vibe Code Bench 53.5% #63 of 120 Official leaderboard
SWE-bench Verified 73.8% #17 of 32 Epoch AI run
PaperBench 39% #8 of 15 Lab model card
DesignArena UI Component 1,181 #78 of 139 Official leaderboard
SWE-Bench Pro Private 23.8% #7 of 12 Official leaderboard
TasteVal 27.7× #12 of 20 Official leaderboard
DesignArena Data Visualization 1,184 #72 of 118 Official leaderboard
Code Arena 1,420 #79 of 128 Official leaderboard
DesignArena Code 1,160 #90 of 143 Official leaderboard
DesignArena SVG 1,129 #66 of 103 Official leaderboard
FrontierCS 32.4 #9 of 14 Official leaderboard
SWE-Bench Pro 29.9% #16 of 24 Official leaderboard
OpenAI-Proof Q&A 3.00% #6 of 9 Lab model card
SWE-bench Multilingual 66.7% #9 of 13 Official leaderboard
MirrorCode 32.2% #28 of 39 Epoch AI run
DesignArena 3D Design 1,062 #105 of 136 Official leaderboard
ts-bench v1 92% #7 of 9 Official leaderboard
DesignArena Agentic Overall 1,057 #59 of 74 Official leaderboard
DesignArena Agentic Image to HTML 1,153 #9 of 11 Official leaderboard
DesignArena React Native 1,088 #47 of 56 Official leaderboard
DesignArena Android (Kotlin) 1,045 #40 of 46 Official leaderboard
DesignArena Full Stack 1,021 #43 of 49 Official leaderboard
Vals RSI Index 0.16 #24 of 27 Official leaderboard
DesignArena Frontend Web App 1,067 #48 of 53 Official leaderboard
WeirdML v3 5.01% #25 of 27 Official leaderboard
Data Agent Benchmark 29.9% #18 of 18 Official leaderboard
DataPrep-Bench Finance 48.1 #5 of 5 Official leaderboard

Agents and tools

BenchmarkScoreRankSource
Elimination Game 8 #1 of 56 Official leaderboard
MCPMark 57.5% #1 of 28 Official leaderboard
Agent Poker Bench 1,132 #1 of 17 Official leaderboard
Pencil Puzzle Bench 56% #4 of 53 Official leaderboard
GAIA 86% #6 of 64 Official leaderboard
AgencyBench 56.5% #1 of 9 Official leaderboard
SnakeBench 33 #21 of 188 Official leaderboard
ScreenSpot-Pro 86.3% #4 of 32 Lab model card
BalatroBench 18.4 #2 of 16 Official leaderboard
DeepPlanning 48.2% #2 of 15 Official leaderboard
RWSearch 82% #1 of 7 Official leaderboard
WorldTravel 32.7% #1 of 7 Official leaderboard
Terminal-Bench 2.0 64.9% #8 of 43 Official leaderboard
LLM Chess 1,085 #23 of 123 Official leaderboard
tau2-bench 90.3% #15 of 79 Lab model card
BFCL V4 55.9% #13 of 65 Official leaderboard
DeR2 Bench 83.8% #1 of 5 Official leaderboard
Chess Puzzles 48% #20 of 93 Epoch AI run
Clocktower Radio 1,694 #5 of 23 Official leaderboard
LisanBench 3592.3 #32 of 133 Official leaderboard
Game Arena Five in a Row 817 #5 of 20 Official leaderboard
tau2-bench Verified 80.3% #3 of 11 Official leaderboard
Toolathlon 43.8% #14 of 50 Official leaderboard
LLM-WikiRace 10% #6 of 21 Official leaderboard
Game Arena Poker 4.98 #9 of 31 Official leaderboard
VITA-Bench 24.3% #6 of 18 Official leaderboard
Game Arena Chess Openings 889 #4 of 11 Official leaderboard
Game Arena Clobber 348 #15 of 39 Official leaderboard
tau3-bench 84.8% #3 of 7 Official leaderboard
Game Arena Reversi 625 #17 of 39 Official leaderboard
Game Arena Checkers 973 #18 of 41 Official leaderboard
GameWorld 40.6% #10 of 22 Official leaderboard
SkillsBench 51.7% #10 of 22 Official leaderboard
Game Arena Lines of Action 427 #18 of 39 Official leaderboard
Game Arena Chess 827 #20 of 42 Official leaderboard
Game Arena Four in a Row 396 #21 of 43 Official leaderboard
Cresta-Bench 49.5% #7 of 14 Official leaderboard
Mystery Game Puzzles 23% #43 of 84 Epoch AI run
τ³-bench Banking 32.2% #13 of 25 Official leaderboard
Kaggle Game Arena 288 #23 of 42 Official leaderboard
Game Arena Dots and Boxes 878 #22 of 40 Official leaderboard
Game Arena Dark Hex 249 #24 of 41 Official leaderboard
Cooperate to Compete (C2C) 0.35 #3 of 5 Official leaderboard
EBR-bench 23% #19 of 31 Epoch AI run
Game Arena Word Association 188 #25 of 40 Official leaderboard
Game Arena Ultimate Tic-Tac-Toe 242 #25 of 39 Official leaderboard
Game Arena Go 561 #27 of 39 Official leaderboard
MCP Atlas 67.6% #27 of 37 Official leaderboard
GameHorizon 62.8 #28 of 37 Official leaderboard
Game Arena Coin Game 246 #30 of 39 Official leaderboard
Game Arena Nine Men's Morris 512 #30 of 39 Official leaderboard
Game Arena Word Art 243 #30 of 39 Official leaderboard
Deep Research Bench (FutureSearch) 0.41 #18 of 21 Official leaderboard
DeepResearch Bench 41.1% #19 of 22 evals.futuresearch.ai, via Epoch AI
LMArena Search Arena 1,161 #32 of 37 Official leaderboard
Game Arena Bargaining 126 #35 of 39 Official leaderboard

Professional work

BenchmarkScoreRankSource
OccuBench 79.6% #1 of 14 Official leaderboard
TaxEval v2 75.8% #11 of 129 Official leaderboard
Vals Finance Agent v1 52.8% #5 of 47 Official leaderboard
CaseLaw v2 66% #8 of 53 Official leaderboard
HealthBench 63.3% #5 of 31 Lab model card
Vals Index v1 64.5% #6 of 34 Official leaderboard
Vals Finance Agent 58.5% #9 of 49 Official leaderboard
MedCode 49.8% #20 of 96 Official leaderboard
CorpFin v2 65.9% #26 of 119 Official leaderboard
HealthBench Hard 42% #5 of 21 Lab model card
ProfBench 55.1% #18 of 72 Official leaderboard
APEX-Agents 23% #2 of 8 Official leaderboard
SpreadsheetBench 2 26.8% #2 of 8 Official leaderboard
prinzbench 52.0 #7 of 26 Official leaderboard
MortgageTax 67.1% #28 of 87 Official leaderboard
MedScribe 84.4% #35 of 97 Official leaderboard
FoodTruck Bench $28,081 #22 of 59 Official leaderboard
AlphaEval 54.9% #2 of 5 Official leaderboard
LegalBench 82.8% #65 of 138 Official leaderboard
APEX-1 64.7% #34 of 71 Official leaderboard
TaxCalcBench 33.8% #10 of 17 Official leaderboard
EnterpriseBench: CoreCraft 42.6% #31 of 51 Official leaderboard
GDPval-MM 54.5% #4 of 6 Official leaderboard
HealthBench Professional 45.9% #4 of 6 Official leaderboard
GDPval 70.9% #7 of 10 Lab model card
Vending-Bench 2 $3,591 #47 of 67 Official leaderboard
Remote Labor Index 2.50% #14 of 19 Official leaderboard
HealthBench Consensus 94.5% #14 of 17 Lab model card
Vending-Bench Arena $619 #29 of 32 Official leaderboard

Vision, audio and robotics

BenchmarkScoreRankSource
MMMU-Pro 80.4% #4 of 57 Official leaderboard
VPCT 84% #2 of 27 VPCT leaderboard, via Epoch AI
EmoBench-M 66.5% #2 of 15 Official leaderboard
ChartMuseum 77% #3 of 21 Official leaderboard
MATH-Vision (MathVision) 83% #16 of 105 Official leaderboard
CharXiv 88.7% #12 of 75 Lab model card
WorldVQA 28% #4 of 22 Official leaderboard
RoboRewardBench 0.89 #5 of 22 Official leaderboard
ERQA 41.5% #4 of 17 Official leaderboard
VisFactor 33.6% #7 of 26 Official leaderboard
CaP-X (CaP-Bench) 22% #3 of 10 Official leaderboard
CiteVQA 33.7% #5 of 16 Official leaderboard
ZeroBench 18% #24 of 76 Official leaderboard
Math Kangaroo 2025 (MathArena) 86.5% #7 of 22 Official leaderboard
VideoDR (with search) 69% #2 of 6 Official leaderboard
Vision Arena 1,242 #52 of 136 Official leaderboard
VISTA 46.6% #21 of 54 Official leaderboard
MADQA 67.8% #7 of 18 Official leaderboard
DISBench 13.1% #4 of 9 Official leaderboard
VoxelBench Image 1,360 #7 of 13 Official leaderboard
MineBench 1,456 #42 of 75 Official leaderboard
ClockBench 15.6% #21 of 37 Official leaderboard
Furniture Assembly Benchmark 38.3% #20 of 35 Official leaderboard
BVB 65.2 #27 of 45 Official leaderboard
MMOU 40.7% #3 of 5 Official leaderboard
BabyVision 34.4% #17 of 27 Official leaderboard
vCX Benchmark 83.2% #28 of 42 Official leaderboard
MDPBench 68.6 #8 of 12 Official leaderboard
Drone-Bench 51.7% #13 of 19 Official leaderboard
OmniDocBench v1.6 86.6 #6 of 8 Official leaderboard
OmniDocBench v1.5 85.8 #9 of 11 Official leaderboard
VoxelBench 1,406 #46 of 56 Official leaderboard
OCR Arena 1,356 #11 of 11 Official leaderboard

Safety and security

BenchmarkScoreRankSource
LiveSecBench 84.7 #3 of 34 Official leaderboard
Lakera B3 (Backbone Breaker Benchmark) 79.6% #4 of 35 Official leaderboard
Spiral-Bench v1.2 70.2 #3 of 21 Official leaderboard
LABBench2 FigQA2 42.6% #1 of 5 Official leaderboard
KORA 61.4% #9 of 42 Official leaderboard
nocot-bench (NCRI) 85.8 #43 of 185 Official leaderboard
Molecular Biology Capabilities Test 54.8% #22 of 90 Official leaderboard
StrongREJECT 97.5% #5 of 19 Lab model card
Capture the Flag (Professional) 82% #6 of 21 Lab model card
Human Pathogen Capabilities Test 60.1% #25 of 86 Official leaderboard
Human Pathogen Capabilities Test v2 (HPCT v2) 63.2% #22 of 71 Official leaderboard
FORTRESS 17.5% #24 of 73 Official leaderboard
ProtocolQA Open-Ended 40% #9 of 27 Lab model card
BioTIER Refuse 75.4% #27 of 69 Official leaderboard
LABBench2 LitQA3 81.5% #2 of 5 Official leaderboard
LABBench2 SeqQA2 49.5% #2 of 5 Official leaderboard
DTap Direct Misuse 38% #4 of 8 Official leaderboard
CyberGym-E2E (615 tasks) 20.7% #2 of 4 Official leaderboard
Deception in production traffic 1.60% #1 of 2 Lab model card
PlugInject 99.6% #1 of 2 Lab model card
TroubleshootingBench 35% #11 of 20 Lab model card
GeneBench Pro 8.53% #7 of 12 Official leaderboard
BioTIER Permit 99.2% #42 of 69 Official leaderboard
DTap Prompt Injection 35.2% #5 of 8 Official leaderboard
BinaryAudit 17.5% #13 of 20 Official leaderboard
CVE-Bench 69% #11 of 14 Lab model card
First-Person Fairness evaluation 1.00% #8 of 10 Lab model card
DrugDiscoveryBench 29.3% #17 of 20 Official leaderboard
Agent JSK 97.8% #6 of 7 Lab model card
Multimodal Troubleshooting Virology 43% #24 of 27 Lab model card
Broken Search Tool 9.10% #8 of 9 Lab model card
PCT (Polarized Contrastive Pairs) 57% #41 of 46 Official leaderboard
Tacit Knowledge and Troubleshooting 48% #30 of 30 Lab model card
Coding Deception 25.6% #6 of 6 Lab model card
LABBench2 ProtocolQA2 41.6% #5 of 5 Official leaderboard
CharXiv Missing Image 88.8% #2 of 2 Lab model card
Production Deception - Adversarial 5.40% #1 of 1 Lab model card

Other

BenchmarkScoreRankSource
ACCoRD 79.3 #15 of 152 Official leaderboard
Conceptual Reasoning Index 59.2 #41 of 151 Official leaderboard
SAGE (Vals) 49.3% #22 of 81 Official leaderboard
LMCA 43.9% #47 of 147 Official leaderboard
CAIS Text Capabilities Index 33.5 #32 of 68 Official leaderboard
CAIS Risk Index 50.3 #32 of 67 Official leaderboard
CAIS Vision Capabilities Index 52.2 #26 of 53 Official leaderboard
TutorBench 53.5% #10 of 20 Official leaderboard

Its lab's numbers put it at 153.2, 1.3 points below its ECI

The Self-Reported Index fits one score to the 19 capability results OpenAI 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 151.0 to 155.4. How we compute it.

Its lab's numbers next to 8 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 Pro 55.6% SWE-Bench Pro (public) Software engineering 29.9% Leaderboard, SWE-agent +26 pts
SWE-bench Verified 80% SWE-bench Verified Software engineering 72.8% Leaderboard, high, mini-SWE-agent +7.2 pts
FrontierMath Tier 4 14.6% FrontierMath (Tier 4) Advanced mathematics 18.8% Epoch AI run, xhigh −4.2 pts
GPQA Diamond 92.4% GPQA Diamond (no tools) Science questions 91.4% Epoch AI run, xhigh +1.0 pts
FrontierMath 40.3% FrontierMath (Tier 1–3) Advanced mathematics 40.7% Leaderboard, xhigh; task version 1.1.2, Epoch −0.4 pts
ARC-AGI 86.2% ARC-AGI-1 (Verified) Abstract reasoning 86.2% Leaderboard, xhigh Same
ARC-AGI-2 52.9% ARC-AGI-2 (Verified) Abstract reasoning 52.9% Leaderboard, xhigh Same
AIME 2025 100% AIME 2025 (no tools) Competition math 100% Leaderboard, high Same

OpenAI's models over time

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

50 100 150 2023 2024 2025 2026 2027 GPT-6.1 Terra GPT-6 Luna GPT-5.4 Mini GPT-5.2 Pro GPT-5 GPT-5 nano GPT-4.1 nano 50 100 150 2023 2024 2025 2026 2027 GPT-6.1 Terra GPT-6 Luna

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.

Introducing GPT-5.2 Dec 11, 2025

Its system card

Every result in the longer documents the lab published with it. 29 of them aren't in the launch post.

Update to GPT-5 System Card: GPT-5.2 Dec 11, 2025

Capabilities

Health

Hallucination

Safety

Robustness

Bias

Alignment

Bio and chem risk

Cyber risk

AI R&D risk