Introducing Grok 4.5 Jul 16, 2026
Grok 4.5
Grok 4.5 ranks first on 2 of the 189 benchmarks with five or more models scored.
205
Benchmarks with a score
2
Ranked first, out of five or more models
5
In the top three
xAI Jul 2026
Grok 4.5
Closed weights Flagship
155.0
#46 of 256
Epoch Capabilities Index
Self-Reported Index 152.2 #81 of 404
- DeepSWE 53.8% #19 /28
- Agents' Last Exam 27% #12 /31
- Terminal-Bench 4.0 12.4% #32 /40
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 | |
|---|---|---|---|---|
| LMArena (Elo) | 1,466 | #44 of 335 | Official leaderboard | |
| BullshitBench | 60% | #18 of 107 | Official leaderboard | |
| SimpleQA Verified | 53.8% | #13 of 56 | Official leaderboard | |
| Judgemark v4 | 77.1 | #10 of 40 | Official leaderboard | |
| DesignArena ASCII Art | 1,249 | #20 of 77 | Official leaderboard | |
| Global MMLU Lite | 93.3% | #13 of 43 | Official leaderboard | |
| BullshitBench V2 | 55% | #44 of 129 | Official leaderboard | |
| StudyArena Text | 1,240 | #34 of 85 | Official leaderboard | |
| Extended NYT Connections | 79.9% | #43 of 103 | Official leaderboard | |
| Creative Writing v3 | 70.5 | #51 of 120 | Official leaderboard | |
| FACTS Parametric | 46.8% | #22 of 45 | Official leaderboard | |
| xAI Factuality (Hallucination) eval | 0.98% | #1 of 2 | Lab model card | |
| Document Arena | 1,466 | #20 of 39 | Official leaderboard | |
| AA-LCR Extended | 60.4% | #30 of 49 | Official leaderboard | |
| HLE Extended | 25.4% | #20 of 31 | Official leaderboard | |
| FACTS Search V2 | 79.4% | #14 of 21 | Official leaderboard | |
| FACTS Parametric V2 | 42.5% | #20 of 27 | Official leaderboard | |
| Antidote | 973 | #27 of 36 | Official leaderboard | |
| Hemingway-bench | 972 | #35 of 42 | Official leaderboard | |
| Lech Mazur Creative Writing | -5.13 | #50 of 50 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DTBench | 96.5% | #13 of 180 | Official leaderboard | |
| GPQA Diamond | 93.4% | #20 of 229 | Epoch AI run | |
| ForecastBench | 0.12 | #9 of 82 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 97.8% | #38 of 212 | Epoch AI run | |
| SimpleBench | 70% | #25 of 106 | Official leaderboard | |
| BixBench | 93.8% | #1 of 4 | Lab model card | |
| xbench-ScienceQA | 72.4% | #9 of 29 | Official leaderboard | |
| Surface Evolver Bench | 74.4% | #8 of 25 | Official leaderboard | |
| ProofBench | 30% | #23 of 59 | Official leaderboard | |
| ARC-AGI | 87.2% | #42 of 95 | Official leaderboard | |
| ARC-AGI-2 | 52.6% | #44 of 98 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 57.2% | #46 of 86 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 24.4% | #43 of 68 | Official leaderboard | |
| ProofBench v1.1 | 31% | #37 of 52 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE Atlas | 84% | #1 of 14 | Lab model card | |
| SWE-Marathon | 29% | #1 of 12 | Official leaderboard | |
| APEX-SWE | 53.6% | #9 of 65 | Official leaderboard | |
| SWE-rebench | 63.8% | #2 of 13 | Official leaderboard | |
| GBAEval | 65.4% | #4 of 21 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,227 | #9 of 46 | Official leaderboard | |
| SWE-Bench Pro | 64.7% | #16 of 81 | Lab model card | |
| SWE-bench Verified Extended | 70.7% | #9 of 44 | Official leaderboard | |
| DesignArena Website | 1,274 | #34 of 151 | Official leaderboard | |
| FrogsGame post-training task | 3.40% | #4 of 17 | Official leaderboard | |
| FrontierSWE v1 | 72.1% | #4 of 17 | Official leaderboard | |
| DesignArena Agentic Game Dev | 1,184 | #7 of 27 | Official leaderboard | |
| DesignArena Data Visualization | 1,265 | #32 of 118 | Official leaderboard | |
| SWE-bench Multilingual | 78% | #20 of 73 | Lab model card | |
| DesignArena Full Stack | 1,227 | #14 of 49 | Official leaderboard | |
| DesignArena Code | 1,265 | #41 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,270 | #40 of 139 | Official leaderboard | |
| WebDev Arena | 1,555 | #31 of 105 | arena.ai, via Epoch AI | |
| DesignArena 3D Design | 1,238 | #43 of 136 | Official leaderboard | |
| DesignArena Game Dev | 1,252 | #46 of 143 | Official leaderboard | |
| DesignArena React Native | 1,208 | #19 of 56 | Official leaderboard | |
| Code Arena | 1,552 | #45 of 128 | Official leaderboard | |
| DesignArena SVG | 1,201 | #37 of 103 | Official leaderboard | |
| DesignArena Agentic Overall | 1,213 | #27 of 74 | Official leaderboard | |
| Vibe Code Bench | 69% | #52 of 120 | Official leaderboard | |
| Code Migration | 36.6% | #35 of 78 | Official leaderboard | |
| Image-to-WebDev Arena | 1,562 | #25 of 55 | Official leaderboard | |
| DesignArena Frontend Web App | 1,199 | #25 of 53 | Official leaderboard | |
| ALE-Bench | 1,467 | #61 of 126 | Official leaderboard | |
| WeirdML | 46.4% | #55 of 112 | Official leaderboard | |
| DesignArena Godot Game Dev | 1,209 | #12 of 24 | Official leaderboard | |
| DesignArena Image to Frontend | 1,191 | #17 of 31 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 12% | #21 of 38 | Official leaderboard | |
| DesignArena Image to HTML | 1,194 | #30 of 51 | Official leaderboard | |
| ReactBench | 32.3% | #17 of 28 | Official leaderboard | |
| Senior SWE-Bench | 23.2% | #14 of 23 | Official leaderboard | |
| DeepSWE | 53.8% | #19 of 28 | Official leaderboard | |
| Chart Arena | 1,466 | #11 of 15 | Official leaderboard | |
| MirrorCode | 32.2% | #29 of 39 | Epoch AI run | |
| IOI v2 (Vals) | 40.6% | #34 of 45 | Official leaderboard | |
| PostTrainBench 1.1 | 23.4% | #9 of 11 | Official leaderboard | |
| ProgramBench tests passed | 57.2% | #9 of 10 | Lab model card | |
| PostTrainBench 1.2 | 22.6% | #19 of 21 | Official leaderboard | |
| RelBench | 38.4% | #1 of 1 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| τ³-bench Banking | 47.9% | #3 of 25 | Official leaderboard | |
| Long-Horizon-Terminal-Bench (LHTB) | 50.5% | #4 of 25 | Official leaderboard | |
| WildClawBench | 57.5% | #6 of 31 | Official leaderboard | |
| RuneBench | 5.70 | #18 of 76 | Official leaderboard | |
| Game Arena Bargaining | 218 | #11 of 39 | Official leaderboard | |
| Terminal-Bench 2.1 | 79.3% | #4 of 13 | Official leaderboard | |
| Game Arena Lines of Action | 469 | #14 of 39 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 377 | #15 of 39 | Official leaderboard | |
| LMArena Search Arena | 1,205 | #15 of 37 | Official leaderboard | |
| PinchBench 2.0 | 75.2% | #25 of 57 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 33.3% | #21 of 44 | Official leaderboard | |
| Game Arena Coin Game | 363 | #19 of 39 | Official leaderboard | |
| Bench on the Clocktower | 50% | #5 of 10 | Official leaderboard | |
| Game Arena Four in a Row | 368 | #22 of 43 | Official leaderboard | |
| Chess Puzzles | 35% | #48 of 93 | Epoch AI run | |
| Game Arena Go | 904 | #21 of 39 | Official leaderboard | |
| Game Arena Word Association | 254 | #22 of 40 | Official leaderboard | |
| Game Arena Poker | -0.85 | #18 of 31 | Official leaderboard | |
| BabaIsBench | 75% | #19 of 32 | Official leaderboard | |
| Game Arena Dots and Boxes | 857 | #24 of 40 | Official leaderboard | |
| DuelLab GameBench 2 | 61.5 | #35 of 57 | Official leaderboard | |
| Kaggle Game Arena | 263 | #27 of 42 | Official leaderboard | |
| Game Arena Chess | 645 | #29 of 42 | Official leaderboard | |
| Game Arena Nine Men's Morris | 530 | #27 of 39 | Official leaderboard | |
| Game Arena Dark Hex | 173 | #29 of 41 | Official leaderboard | |
| Terminal-Bench 3.0 | 15.7% | #8 of 11 | Official leaderboard | |
| Game Arena Clobber | 153 | #29 of 39 | Official leaderboard | |
| Game Arena Reversi | 240 | #29 of 39 | Official leaderboard | |
| BrowseComp Extended | 14.2% | #32 of 42 | Official leaderboard | |
| CEO-Bench | $15,909 | #15 of 19 | Official leaderboard | |
| Terminal-Bench 4.1 | 12.4% | #34 of 43 | Official leaderboard | |
| Terminal-Bench 4.0 | 12.4% | #32 of 40 | Official leaderboard | |
| Game Arena Word Art | 177 | #33 of 39 | Official leaderboard | |
| Time Horizon Index: KSP | 6.33% | #11 of 13 | Official leaderboard | |
| ARC-AGI-3 | 0.32% | #23 of 27 | Official leaderboard | |
| Game Arena Checkers | 83 | #39 of 41 | Official leaderboard | |
| DeepSearchQA | 38.4% | #22 of 23 | Lab model card | |
| tau3-bench | 32.6% | #14 of 14 | Lab model card | |
| Vals Web Search Index | 38.8% | #5 of 5 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| YC-Bench | $2,629,903 | #5 of 59 | Official leaderboard | |
| CorpFin v2 | 67.4% | #13 of 119 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 12.9% | #11 of 80 | Official leaderboard | |
| LegalBench | 86% | #19 of 138 | Official leaderboard | |
| MedScribe | 86.9% | #19 of 97 | Official leaderboard | |
| FoodTruck Bench | $34,586 | #20 of 59 | Official leaderboard | |
| TRLawBench | 9.35 | #12 of 34 | Official leaderboard | |
| APEX-Agents 1.1 | 70.5% | #23 of 65 | Official leaderboard | |
| ProfBench | 51% | #27 of 72 | Official leaderboard | |
| APEX-Agents | 56.2% | #23 of 61 | Collected by Epoch AI, source not named | |
| Agents' Last Exam | 27% | #12 of 31 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,195 | #15 of 36 | Official leaderboard | |
| RadLE 2.0 | 524.0 | #5 of 12 | Official leaderboard | |
| DesignArena HTML Slides | 1,100 | #12 of 28 | Official leaderboard | |
| PolicyBench | 84.7% | #22 of 50 | Official leaderboard | |
| MedCode | 43.3% | #44 of 96 | Official leaderboard | |
| Legal Research Bench | 38% | #44 of 93 | Official leaderboard | |
| Vals Index v2 | 51.5% | #35 of 65 | Official leaderboard | |
| APEX-Accounting | 57.4% | #27 of 49 | Official leaderboard | |
| GDPval-AA v2 | 1,535 | #15 of 27 | Lab model card | |
| TaxEval v2 | 71.7% | #72 of 129 | Official leaderboard | |
| Tax Agent Bench | 24.9% | #47 of 84 | Official leaderboard | |
| EEBench | 50.9% | #18 of 30 | Official leaderboard | |
| Vals Finance Agent v2 | 48.4% | #50 of 80 | Official leaderboard | |
| Vending-Bench 2 | $3,887 | #42 of 67 | Official leaderboard | |
| MortgageTax | 61.8% | #55 of 87 | Official leaderboard | |
| APEX-1 | 61.8% | #46 of 71 | Official leaderboard | |
| Excel Modeling Benchmark | 52.9% | #52 of 76 | Official leaderboard | |
| DesignArena Spreadsheet | 1,091 | #26 of 35 | Official leaderboard | |
| GDPval Extended | 0.40% | #32 of 43 | Official leaderboard | |
| DesignArena PDF | 1,137 | #29 of 37 | Official leaderboard | |
| Vals Index v2.1 | 44.7% | #40 of 49 | Official leaderboard | |
| $OneMillion-Bench | 15.8% | #12 of 13 | Official leaderboard | |
| MedXpertQA MM Extended | 41.2% | #40 of 41 | Official leaderboard | |
| HW-Bench | 3.00% | #15 of 15 | Official leaderboard | |
| 3DCodeBench | 49.8% | #3 of 3 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vision Arena | 1,280 | #21 of 136 | Official leaderboard | |
| ZeroBench | 15% | #29 of 76 | Official leaderboard | |
| ClockBench | 21.7% | #18 of 37 | Official leaderboard | |
| VoxelBench | 1,741 | #28 of 56 | Official leaderboard | |
| MineBench | 1,538 | #38 of 75 | Official leaderboard | |
| BVB | 66.4 | #24 of 45 | Official leaderboard | |
| DesignArena Code to Video | 1,123 | #17 of 29 | Official leaderboard | |
| Chartography | 17.3% | #26 of 43 | Official leaderboard | |
| Blueprint-Bench 2 | 27.3% | #25 of 38 | Official leaderboard | |
| MMMU-Pro Extended | 29.8% | #27 of 38 | Official leaderboard | |
| PerceptionBench (Moonshot) | 41% | #12 of 16 | Official leaderboard | |
| ArtifactArena | 957 | #18 of 21 | Official leaderboard | |
| CharXiv Extended | 72.7% | #33 of 38 | Official leaderboard | |
| Furniture Assembly Benchmark | 22.5% | #32 of 35 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Virology Capabilities Test (VCT) | 65.5% | #2 of 31 | Lab model card | |
| ProtocolQA Open-Ended (xAI) | 87% | #1 of 4 | Lab model card | |
| xAI sycophancy eval | 0.01% | #1 of 4 | Lab model card | |
| nocot-bench (NCRI) | 81.0 | #57 of 185 | Official leaderboard | |
| dfbench | 57.2% | #6 of 18 | Official leaderboard | |
| Child safety (xAI) | 0% | #1 of 3 | Lab model card | |
| FORTRESS-RN | 97.9% | #1 of 3 | Lab model card | |
| HackerBench | 7.80% | #1 of 3 | Lab model card | |
| Self-harm refusals (xAI) | 0.50% | #1 of 3 | Lab model card | |
| BioTIER Permit | 99.7% | #24 of 69 | Official leaderboard | |
| BioTIER Refuse | 76.2% | #25 of 69 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 61.2% | #26 of 71 | Official leaderboard | |
| CyberGym | 79% | #14 of 34 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 71% | #19 of 44 | Official leaderboard | |
| SAFE-Teen Health advice | 47% | #9 of 17 | Official leaderboard | |
| Dictatorship Eval | 66% | #11 of 20 | Official leaderboard | |
| SecIT Bench | 74.9% | #15 of 26 | Official leaderboard | |
| DeepsecBench | 16.4 | #19 of 29 | Official leaderboard | |
| VCT-v2 | 41.2% | #17 of 25 | Official leaderboard | |
| SAFE-Teen Urgent safety | 50% | #12 of 17 | Official leaderboard | |
| PCT (Polarized Contrastive Pairs) | 50.2% | #34 of 46 | Official leaderboard | |
| General refusals (xAI) | 1.10% | #6 of 8 | Lab model card | |
| SRE-Bench | 0.76% | #27 of 35 | Official leaderboard | |
| SAFE-Teen Relationship boundaries | 65% | #15 of 17 | Official leaderboard | |
| Autointent-Bio | 97.9% | #3 of 3 | Lab model card | |
| Autointent-Chem | 96.7% | #3 of 3 | Lab model card | |
| LAB-Bench | 71.1% | #3 of 3 | Lab model card | |
| Standard jailbreaks (xAI) | 0.73% | #3 of 3 | Lab model card | |
| Epistemic Bias (xAI) | 20.4% | #1 of 1 | Lab model card | |
| FalseClaimBench | 84% | #1 of 1 | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Conceptual Reasoning Index | 60.3 | #34 of 151 | Official leaderboard | |
| ACCoRD | 71.6 | #39 of 152 | Official leaderboard | |
| LMCA | 45.2% | #42 of 147 | Official leaderboard | |
| CAIS Vision Capabilities Index | 56.2 | #21 of 53 | Official leaderboard | |
| Vals Multimodal Index | 63.4% | #15 of 33 | Official leaderboard | |
| CAIS Text Capabilities Index | 32.3 | #34 of 68 | Official leaderboard | |
| CAIS Risk Index | 53.4 | #38 of 67 | Official leaderboard | |
| SAGE (Vals) | 35% | #68 of 81 | Official leaderboard |
Its lab's numbers put it at 152.2, 2.8 points below its ECI
The Self-Reported Index fits one score to the 10 capability results xAI 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 144.3 to 160.0. How we compute it.
- SWE Atlas 84
- Terminal-Bench 2.1 83.3
- SWE-bench Multilingual 78
- SWE-Bench Pro 64.7
- DeepSWE 62
- ProgramBench tests passed 57.2
- DeepSearchQA 38.4
- tau3-bench 32.6
- τ³-bench Banking 32.6
- SWE-Marathon 29
Its lab's numbers next to 7 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 |
|---|---|---|---|---|
| τ³-bench Banking | 32.6% τ³-banking (AA), accuracy, high (4.2) | 47.9% Leaderboard, high | −15 pts | xAI quotes Artificial Analysis's run of the same tasks, with its own user simulator and judge. The board shows Sierra's own runs. |
| DeepSWE | 62% Best of DeepSWE 1.0 (pass@1; eval by Datacurve, run with provider harnesses by AA) / DeepSWE 1.1 (mini-swe-agent harness run by Datacurve) | 53.8% Leaderboard, high, mini-swe-agent | +8.2 pts | |
| FrontierSWE v1 | 78% FrontierSWE, Dominance (win rate versus a random baseline), high, Grok CLI, public leaderboard (2.7) | 72.1% Leaderboard, mean, Grok CLI | +5.9 pts | |
| Terminal-Bench 2.1 | 83.3% Terminal Bench 2.1 | 79.3% Leaderboard, high, Cursor CLI | +4.0 pts | |
| APEX-SWE | 51.2% APEX-SWE, Pass@1, high, run by Mercor (2.3) | 53.6% Leaderboard, high, terminus-2 | −2.4 pts | |
| CyberGym | 80.4% CyberGym, mean reproduced, high, unrestricted (7.1) | 79% Leaderboard, Grok Build | +1.4 pts | |
| SWE-Marathon | 29% SWE Marathon resolution rate (pass@1) | 29% Leaderboard, Grok Build | Same |
xAI's models over time
xAI'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.
Its system card
Every result in the longer documents the lab published with it. 32 of them aren't in the launch post.
Model Card: Grok 4.5 Jul 14, 2026
Capabilities
- DeepSWE 62
- APEX-SWE 51.2
- SWE-Bench Pro 64.7
- SWE-bench Multilingual 78
- SWE-Marathon 29
- FrontierSWE v1 78
- ProgramBench tests passed 57.2
- Terminal-Bench 2.1 83.3
- SWE Atlas 84
- EEBench
- 3DCodeBench 49.8
- Parametric CAD Bench
- GDPval-AA v2 1535
- τ³-bench Banking 32.6
- DeepSearchQA 38.4
Hallucination
Safety
- General refusals (xAI) 1.1
- Child safety (xAI) 0
- Autointent-Bio 97.9
- Autointent-Chem 96.7
- FORTRESS-RN 97.9
- Self-harm refusals (xAI) 0.5
Robustness
Bias
- Epistemic Bias (xAI) 20.4
Alignment
- FalseClaimBench 84
- MASK
- xAI sycophancy eval 0.01
Bio and chem risk
- Virology Capabilities Test (VCT) 65.5
- WMDP
- LAB-Bench 71.1
- ProtocolQA Open-Ended (xAI) 87
- BixBench 93.8
Cyber risk
- CyberGym 80.4
- HackerBench 7.8