Introducing Grok 4.6 Aug 12, 2026
- AA Intelligence Index 61
- GDPVal-AA v2 1753
- CursorBench v3.2 69.9
- DeepSWE v1.1 65.9
- FrontierCode v1.1 (Extended) 61.3
- APEX-Agents 57.5
- Terminal-Bench v3.0 26
- APEX-SWE 56.4
- AA-Briefcase 1577
- Harvey LAB (Vals) 15.8
Grok 4.6
Grok 4.6 ranks first on 4 of the 253 benchmarks with five or more models scored.
276
Benchmarks with a score
4
Ranked first, out of five or more models
17
In the top three
xAI Aug 2026
Grok 4.6
Closed weights Flagship
157.8
#28 of 256
Epoch Capabilities Index
Self-Reported Index 160.7 #26 of 404
- DeepSWE 67.5% #9 /28
- Terminal-Bench 4.0 20.3% #25 /40
- ZeroBench 17% #26 /76
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 | |
|---|---|---|---|---|
| BullshitBench | 63.6% | #15 of 107 | Official leaderboard | |
| LMArena (Elo) | 1,456 | #54 of 335 | Official leaderboard | |
| DesignArena ASCII Art | 1,267 | #15 of 77 | Official leaderboard | |
| BullshitBench V2 | 66% | #27 of 129 | Official leaderboard | |
| SimpleQA Verified | 52.9% | #15 of 56 | Official leaderboard | |
| HLE Extended | 45.8% | #11 of 31 | Official leaderboard | |
| OpenAI MRCR v2 | 47.7% | #18 of 47 | Official leaderboard | |
| Extended NYT Connections | 80% | #42 of 103 | Official leaderboard | |
| AA-LCR Extended | 65.2% | #22 of 49 | Official leaderboard | |
| FACTS Grounding v2 | 66.5% | #14 of 25 | Official leaderboard | |
| FACTS Parametric V2 | 44.5% | #18 of 27 | Official leaderboard | |
| StudyArena Text | 1,210 | #62 of 85 | Official leaderboard | |
| FACTS Benchmark Suite | 59.5% | #17 of 23 | Official leaderboard | |
| Document Arena | 1,450 | #29 of 39 | Official leaderboard | |
| Antidote | 966 | #28 of 36 | Official leaderboard | |
| Hemingway-bench | 983 | #33 of 42 | Official leaderboard | |
| FACTS Multimodal | 39.3% | #19 of 23 | Official leaderboard | |
| Joke Arena | 1,469 | #21 of 25 | Official leaderboard | |
| Lech Mazur Creative Writing | -2.93 | #46 of 50 | Official leaderboard | |
| xAI Factuality (Hallucination) eval | 1.70% | #2 of 2 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DTBench | 97.3% | #7 of 180 | Official leaderboard | |
| GPQA Diamond | 94% | #15 of 229 | Epoch AI run | |
| OTIS Mock AIME 2024-2025 | 99.2% | #22 of 212 | Epoch AI run | |
| ForecastBench | 0.12 | #10 of 82 | Official leaderboard | |
| SimpleBench | 75.9% | #17 of 106 | Official leaderboard | |
| FrontierChallenge | 20.6% | #2 of 11 | Official leaderboard | |
| BixBench3 | 48.8 | #5 of 20 | Official leaderboard | |
| ProofBench | 45% | #17 of 59 | Official leaderboard | |
| ARC-AGI-2 | 67.1% | #29 of 98 | Official leaderboard | |
| LatchBio Capabilities v1.0 | 47.4% | #8 of 27 | Official leaderboard | |
| scBench (LatchBio SingleCellBench) | 59.1% | #8 of 26 | Official leaderboard | |
| BioMysteryBench | 72.2% | #12 of 30 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 66% | #35 of 86 | Official leaderboard | |
| ARC-AGI | 87.5% | #40 of 95 | Official leaderboard | |
| MysteryMechanism | 30.6% | #14 of 31 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 31.7% | #32 of 68 | Official leaderboard | |
| ExplorationBench | 69% | #5 of 10 | Official leaderboard | |
| BixBench | 93.8% | #2 of 4 | Lab model card | |
| ProofBench v1.1 | 51% | #30 of 52 | Official leaderboard | |
| Prophet Arena | 85.6 | #14 of 24 | Official leaderboard | |
| xbench-ScienceQA | 70.4% | #19 of 29 | Official leaderboard | |
| SciUniverse | 26.2% | #4 of 6 | Official leaderboard | |
| FrontierPhysics | 2.40% | #18 of 22 | Official leaderboard | |
| Terminal-Bench-Science | 7.14% | #14 of 17 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DesignArena Android (Kotlin) | 1,280 | #3 of 46 | Official leaderboard | |
| APEX-SWE | 57.7% | #6 of 65 | Official leaderboard | |
| WebDev Arena | 1,618 | #15 of 105 | arena.ai, via Epoch AI | |
| FrontierCode 1.1 | 61.3% | #2 of 13 | Lab model card | |
| SWE-bench Verified Extended | 78% | #7 of 44 | Official leaderboard | |
| FrogsGame post-training task | 9.30% | #3 of 17 | Official leaderboard | |
| FrontierSWE v1 | 77.9% | #3 of 17 | Official leaderboard | |
| DesignArena Data Visualization | 1,282 | #22 of 118 | Official leaderboard | |
| DesignArena Full Stack | 1,242 | #10 of 49 | Official leaderboard | |
| WeirdML | 67.3% | #23 of 112 | Official leaderboard | |
| DesignArena Godot Game Dev | 1,256 | #5 of 24 | Official leaderboard | |
| DesignArena Code | 1,281 | #30 of 143 | Official leaderboard | |
| DesignArena Website | 1,278 | #32 of 151 | Official leaderboard | |
| Code Arena | 1,618 | #28 of 128 | Official leaderboard | |
| DesignArena UI Component | 1,285 | #31 of 139 | Official leaderboard | |
| DesignArena Game Dev | 1,282 | #32 of 143 | Official leaderboard | |
| Image-to-WebDev Arena | 1,607 | #13 of 55 | Official leaderboard | |
| DesignArena React Native | 1,226 | #14 of 56 | Official leaderboard | |
| Code Migration | 44.6% | #20 of 78 | Official leaderboard | |
| DesignArena 3D Design | 1,257 | #35 of 136 | Official leaderboard | |
| ALE-Bench | 1,822 | #34 of 126 | Official leaderboard | |
| DesignArena Agentic Overall | 1,230 | #21 of 74 | Official leaderboard | |
| DesignArena SVG | 1,221 | #30 of 103 | Official leaderboard | |
| DesignArena Frontend Web App | 1,230 | #16 of 53 | Official leaderboard | |
| DAXBench v2 | 83.9% | #15 of 47 | Official leaderboard | |
| DeepSWE | 67.5% | #9 of 28 | Official leaderboard | |
| Vibe Code Bench | 76.2% | #42 of 120 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 19.3% | #14 of 38 | Official leaderboard | |
| Vals RSI Index | 0.26 | #10 of 27 | Official leaderboard | |
| DesignArena Image to HTML | 1,218 | #19 of 51 | Official leaderboard | |
| ReactBench | 33.9% | #11 of 28 | Official leaderboard | |
| DesignArena Agentic Game Dev | 1,158 | #11 of 27 | Official leaderboard | |
| DesignArena Image to Frontend | 1,205 | #14 of 31 | Official leaderboard | |
| Snorkel Agentic Coding 2.0 | 33.6% | #8 of 16 | Official leaderboard | |
| MacroscopeBench | 75.9% | #9 of 16 | Official leaderboard | |
| MirrorCode | 41.1% | #22 of 39 | Epoch AI run | |
| Senior SWE-Bench | 26.3% | #13 of 23 | Official leaderboard | |
| SWE-Together | 60.6% | #13 of 22 | Official leaderboard | |
| SWE-bench CLI | 5.70% | #10 of 16 | Official leaderboard | |
| IOI v2 (Vals) | 47.6% | #29 of 45 | Official leaderboard | |
| RSI-Exam | 36.7% | #12 of 18 | Official leaderboard | |
| AutoResearchExam | 0.49 | #8 of 12 | Official leaderboard | |
| FrontierSWE | 25.3% | #17 of 25 | Official leaderboard | |
| CursorBench | 41.4% | #15 of 21 | cursor.com, via Epoch AI | |
| SWE-Marathon | 31.9% | #8 of 11 | Lab model card | |
| DesignArena Blender | 1,149 | #9 of 12 | Official leaderboard | |
| LibraryDesignBench | 39.7 | #6 of 8 | Official leaderboard | |
| CursorBench 4.0 | 41.4% | #16 of 21 | Official leaderboard | |
| Vibe Code Bench 1-100 | 14.8% | #20 of 26 | Official leaderboard | |
| Landing Page Arena | 1,402 | #12 of 14 | Official leaderboard | |
| Brokk Power Ranking | 1.50 | #19 of 19 | Official leaderboard | |
| CursorBench 3.2 | 70.8% | #2 of 2 | Lab model card | |
| InferenceEval | 46.9% | #1 of 1 | Lab model card | |
| KernelBenchInternal | 37.2% | #1 of 1 | Lab model card | |
| SpaceXAI MTS Eval | 61.1% | #1 of 1 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Terminal-Bench 2.1 Extended | 45.5% | #5 of 44 | Official leaderboard | |
| RuneBench | 6.28 | #9 of 76 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 500 | #6 of 39 | Official leaderboard | |
| TraceTrial-Bench | 76.7% | #2 of 9 | Official leaderboard | |
| Game Arena Lines of Action | 521 | #12 of 39 | Official leaderboard | |
| CEO-Bench | $246,419 | #6 of 19 | Official leaderboard | |
| Game Arena Dark Hex | 295 | #13 of 41 | Official leaderboard | |
| Game Arena Four in a Row | 502 | #14 of 43 | Official leaderboard | |
| Mystery Game Puzzles | 34% | #29 of 84 | Epoch AI run | |
| LLM Chess | 443 | #44 of 123 | Official leaderboard | |
| BabaIsBench | 96% | #12 of 32 | Official leaderboard | |
| Game Arena Bargaining | 214 | #15 of 39 | Official leaderboard | |
| Game Arena Dots and Boxes | 935 | #16 of 40 | Official leaderboard | |
| RecreationBench | 36.7% | #4 of 10 | Official leaderboard | |
| DuelLab GameBench 2 | 68.7 | #23 of 57 | Official leaderboard | |
| Game Arena Go | 1,023 | #16 of 39 | Official leaderboard | |
| Game Arena Word Association | 289 | #17 of 40 | Official leaderboard | |
| Game Arena Nine Men's Morris | 686 | #17 of 39 | Official leaderboard | |
| Terminal-Bench 3.0 | 26.5% | #5 of 11 | Official leaderboard | |
| Game Arena Coin Game | 378 | #18 of 39 | Official leaderboard | |
| Kaggle Game Arena | 324 | #20 of 42 | Official leaderboard | |
| Game Arena Poker | 0.18 | #15 of 31 | Official leaderboard | |
| Game Arena Chess | 808 | #21 of 42 | Official leaderboard | |
| EBR-bench | 30.5% | #16 of 31 | Epoch AI run | |
| Chess Puzzles | 33% | #50 of 93 | Epoch AI run | |
| ARC-AGI-3 | 2.11% | #15 of 27 | Official leaderboard | |
| Agent Arena | -0.19% | #34 of 56 | Official leaderboard | |
| DeepSearchQA | 81.6% | #14 of 23 | Lab model card | |
| BrowseComp Extended | 21.5% | #26 of 42 | Official leaderboard | |
| Terminal-Bench 4.0 | 20.3% | #25 of 40 | Official leaderboard | |
| Terminal-Bench 4.1 | 19.7% | #27 of 43 | Official leaderboard | |
| Game Arena Reversi | 367 | #25 of 39 | Official leaderboard | |
| Game Arena Word Art | 331 | #25 of 39 | Official leaderboard | |
| Game Arena Clobber | 167 | #27 of 39 | Official leaderboard | |
| ApprenticeBench | 13% | #13 of 18 | Official leaderboard | |
| Game Arena Checkers | 552 | #34 of 41 | Official leaderboard | |
| Time Horizon Index: KSP | 5.83% | #12 of 13 | Official leaderboard | |
| Long-Horizon-Terminal-Bench 2.0 | 42% | #14 of 15 | Official leaderboard | |
| Brood War Bench | 11.1% | #6 of 6 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnorkelFinance 2.0 | 25.1% | #1 of 16 | Official leaderboard | |
| OR-License | 88.4% | #1 of 11 | Official leaderboard | |
| LegalBench | 86.3% | #14 of 138 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 15.8% | #9 of 80 | Official leaderboard | |
| FoodTruck Bench | $59,537 | #7 of 59 | Official leaderboard | |
| SnorkelLegal | 40.7% | #2 of 16 | Official leaderboard | |
| Legal Research Bench | 48.1% | #14 of 93 | Official leaderboard | |
| APEX-Agents | 57.5% | #2 of 12 | Lab model card | |
| APEX-Agents 1.1 | 77.3% | #11 of 65 | Official leaderboard | |
| CorpFin v2 | 66.2% | #21 of 119 | Official leaderboard | |
| EnterpriseBench: CoreCraft | 65.6% | #9 of 51 | Official leaderboard | |
| SnorkelRevOps | 15.8% | #3 of 16 | Official leaderboard | |
| WorkplaceAgents | 17.1% | #3 of 14 | Official leaderboard | |
| GDPval-AA v2 | 1,753 | #6 of 27 | Lab model card | |
| Vending-Bench 2 | $9,047 | #15 of 67 | Official leaderboard | |
| Tax Agent Bench | 33.7% | #19 of 84 | Official leaderboard | |
| MedScribe | 86.5% | #22 of 97 | Official leaderboard | |
| DesignArena Spreadsheet | 1,274 | #8 of 35 | Official leaderboard | |
| DesignArena PDF | 1,239 | #9 of 37 | Official leaderboard | |
| SnorkelManufacturing | 16.4% | #4 of 16 | Official leaderboard | |
| DAYJOB: Finance | 21% | #10 of 39 | Official leaderboard | |
| DAYJOB: Healthcare | 14.8% | #10 of 37 | Official leaderboard | |
| APEX-1 | 67.7% | #20 of 71 | Official leaderboard | |
| DesignArena HTML Slides | 1,144 | #8 of 28 | Official leaderboard | |
| Vals Index v2 | 59.2% | #19 of 65 | Official leaderboard | |
| Tuesday Work Index | 60.8 | #13 of 44 | Official leaderboard | |
| GDPval Extended | 1.90% | #13 of 43 | Official leaderboard | |
| YC-Bench | $1,568,241 | #18 of 59 | Official leaderboard | |
| TRLawBench | 9.38 | #11 of 34 | Official leaderboard | |
| Public Benefits Bench | 66.8% | #17 of 50 | Official leaderboard | |
| ATLAS-Finance | 4.67% | #4 of 11 | Official leaderboard | |
| ATLAS-Finance v1.1.0 | 4.67% | #4 of 11 | Official leaderboard | |
| Clipping Bench | $805 | #7 of 19 | Official leaderboard | |
| APEX-Accounting | 60.3% | #19 of 49 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,197 | #14 of 36 | Official leaderboard | |
| MedCode | 44.7% | #38 of 96 | Official leaderboard | |
| Vals Finance Agent v2 | 53.7% | #32 of 80 | Official leaderboard | |
| PolicyBench | 85.8% | #20 of 50 | Official leaderboard | |
| EEBench | 57.1% | #12 of 30 | Official leaderboard | |
| Excel Modeling Benchmark | 62.7% | #31 of 76 | Official leaderboard | |
| AA-Briefcase | 1,577 | #3 of 7 | Lab model card | |
| Vals Index v2.1 | 52.1% | #23 of 49 | Official leaderboard | |
| LifeSci Diamond | 3.60% | #9 of 18 | Official leaderboard | |
| Digital Work Index | 20.5 | #8 of 16 | Official leaderboard | |
| Epoch Automation Reports | 59.1% | #3 of 6 | Official leaderboard | |
| MortgageTax | 64.2% | #44 of 87 | Official leaderboard | |
| PharmaBench (micro1) | 12.5% | #7 of 13 | Official leaderboard | |
| CAD Arena | 0.35 | #7 of 12 | Official leaderboard | |
| TaxEval v2 | 71.1% | #79 of 129 | Official leaderboard | |
| BlenderWorld | 57.0 | #8 of 13 | Official leaderboard | |
| Immigration Bench | 66% | #16 of 25 | Official leaderboard | |
| FrontierCAD Assemblies | 0.33 | #4 of 6 | Official leaderboard | |
| 3DCodeBench | 54% | #2 of 3 | Lab model card | |
| RealmPathology | 72.1% | #9 of 13 | Official leaderboard | |
| Parametric CAD Bench | 27.7% | #8 of 11 | Official leaderboard | |
| MedXpertQA MM Extended | 48.4% | #32 of 41 | Official leaderboard | |
| MBABench 2.0 | 1.00% | #4 of 5 | Official leaderboard | |
| OfficeQA | 63.2% | #12 of 14 | Lab model card | |
| HW-Bench | 13% | #13 of 15 | Official leaderboard | |
| micro1 Medical Index | 64.1% | #7 of 8 | Official leaderboard | |
| Bedside Bench | 83.3% | #15 of 17 | Official leaderboard | |
| SnorkelUnderwrite 2.0 | 17.8% | #16 of 16 | Official leaderboard | |
| FrontierCAD Parts | 0.26 | #6 of 6 | Official leaderboard | |
| CADGenBench | 40.9% | #2 of 2 | Lab model card | |
| PartBench | 58.5% | #1 of 1 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GlobeBench | 28.1% | #2 of 19 | Official leaderboard | |
| VoxelBench | 2,077 | #11 of 56 | Official leaderboard | |
| BVB | 70.9 | #10 of 45 | Official leaderboard | |
| MineBench | 1,793 | #18 of 75 | Official leaderboard | |
| Vision Arena | 1,263 | #34 of 136 | Official leaderboard | |
| ZeroBench | 17% | #26 of 76 | Official leaderboard | |
| CineRecon | 49.0 | #5 of 12 | Official leaderboard | |
| Blueprint-Bench 2 | 33.2% | #16 of 38 | Official leaderboard | |
| ArtifactArena | 1,136 | #9 of 21 | Official leaderboard | |
| ClockBench | 20% | #19 of 37 | Official leaderboard | |
| DesignArena Code to Video | 1,139 | #15 of 29 | Official leaderboard | |
| Furniture Assembly Benchmark | 40% | #19 of 35 | Official leaderboard | |
| Chartography | 20.5% | #24 of 43 | Official leaderboard | |
| CharXiv Extended | 77.3% | #27 of 38 | Official leaderboard | |
| MMMU-Pro Extended | 26% | #31 of 38 | Official leaderboard | |
| Code2Video Benchmark | 1,450 | #17 of 19 | Official leaderboard | |
| 1H-VideoQA | 60% | #18 of 20 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Virology Capabilities Test (VCT) | 67.4% | #1 of 31 | Lab model card | |
| SecIT Bench | 80.7% | #4 of 26 | Official leaderboard | |
| nocot-bench (NCRI) | 86.8 | #40 of 185 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 61.7% | #7 of 30 | Official leaderboard | |
| Human Pathogen Capabilities Test | 61.6% | #21 of 86 | Official leaderboard | |
| TxBench-Antibody-Discovery | 53% | #8 of 30 | Official leaderboard | |
| BioSecBench-Refusal | 46.9% | #6 of 22 | Official leaderboard | |
| BioSecBench-Function | 44% | #8 of 27 | Official leaderboard | |
| TxBench-Oligo-Discovery | 46% | #10 of 30 | Official leaderboard | |
| BioSecBench-Surveillance | 48% | #9 of 27 | Official leaderboard | |
| Autointent-Bio | 100% | #1 of 3 | Lab model card | |
| Autointent-Chem | 100% | #1 of 3 | Lab model card | |
| LAB-Bench | 80.7% | #1 of 3 | Lab model card | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 62.1% | #24 of 71 | Official leaderboard | |
| CyberGym | 79.7% | #12 of 34 | Official leaderboard | |
| World-Class Bio | 50.3% | #27 of 74 | Official leaderboard | |
| VariantBench | 40.4% | #10 of 27 | Official leaderboard | |
| BioTIER Refuse | 76.1% | #26 of 69 | Official leaderboard | |
| MetagenomicsBench | 42% | #11 of 26 | Official leaderboard | |
| Molecular Biology Capabilities Test | 50.4% | #39 of 90 | Official leaderboard | |
| OncologyBench | 18.7% | #13 of 30 | Official leaderboard | |
| EpiBench | 23.6% | #12 of 27 | Official leaderboard | |
| ProtocolQA Open-Ended (xAI) | 79.6% | #2 of 4 | Lab model card | |
| Biosecurity VCT | 47.8% | #1 of 2 | Lab model card | |
| Biopharma Bench | 54.3% | #7 of 13 | Official leaderboard | |
| scBench-Long | 16.5% | #15 of 27 | Official leaderboard | |
| CWE-bench | 57% | #7 of 12 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 67.7% | #26 of 44 | Official leaderboard | |
| SpatialBench-Long | 27.4% | #16 of 27 | Official leaderboard | |
| General refusals (xAI) | 0.93% | #5 of 8 | Lab model card | |
| BioTIER Permit | 99.2% | #44 of 69 | Official leaderboard | |
| AI Wellbeing Index percent positive experiences | 69.7% | #4 of 6 | Official leaderboard | |
| Child safety (xAI) | 0% | #2 of 3 | Lab model card | |
| FORTRESS-RN | 97.9% | #2 of 3 | Lab model card | |
| HackerBench | 6.90% | #2 of 3 | Lab model card | |
| Self-harm refusals (xAI) | 0.84% | #2 of 3 | Lab model card | |
| Standard jailbreaks (xAI) | 0.04% | #2 of 3 | Lab model card | |
| DeepsecBench | 16.1 | #20 of 29 | Official leaderboard | |
| xAI sycophancy eval | 0.04% | #3 of 4 | Lab model card | |
| CVE-Bench | 39.8% | #12 of 14 | Lab model card | |
| PCT (Polarized Contrastive Pairs) | 55.4% | #40 of 46 | Official leaderboard | |
| WellnessBench | 43% | #12 of 12 | Official leaderboard | |
| BioUseBench | 90.7% | #2 of 2 | Lab model card | |
| Long-horizon jailbreaks (xAI) | 1.00% | #2 of 2 | Lab model card | |
| SecureCodeReview (xAI) | 58.7% | #1 of 1 | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Artificial Analysis Intelligence Index v4.1 | 61.0 | #1 of 8 | Lab model card | |
| Conceptual Reasoning Index | 62.5 | #32 of 151 | Official leaderboard | |
| LMCA | 48.5% | #33 of 147 | Official leaderboard | |
| CAIS Text Capabilities Index | 43.5 | #17 of 68 | Official leaderboard | |
| ACCoRD | 71.6 | #41 of 152 | Official leaderboard | |
| Xbench | 40.7 | #5 of 17 | Official leaderboard | |
| CAIS Risk Index | 48.8 | #28 of 67 | Official leaderboard | |
| CompassRank | 78.0 | #7 of 14 | Official leaderboard | |
| Vals Science Index | 38.1% | #14 of 25 | Official leaderboard | |
| LiveBench | 63.5% | #20 of 31 | Official leaderboard | |
| SAGE (Vals) | 28.9% | #78 of 81 | Official leaderboard |
Its lab's numbers put it at 160.7, 2.9 points above its ECI
The Self-Reported Index fits one score to the 7 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 155.6 to 165.9. How we compute it.
- DeepSearchQA 81.6
- DeepSWE 67
- OfficeQA 63.2
- FrontierCode 1.1 61.3
- APEX-Agents 57.5
- SWE-Marathon 31.9
- Harvey LAB (Legal Agent Benchmark) 15.8
Its lab's numbers next to 6 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 |
|---|---|---|---|
| EEBench | 60% EEBench V1 core corpus, reward, xhigh, Grok Build harness, run by Atopile (4.1). High = 53.0% | 57.1% Leaderboard, Grok CLI | +2.9 pts |
| APEX-SWE | 56.4% APEX-SWE | 57.7% Leaderboard, xhigh, terminus-2 | −1.3 pts |
| Terminal-Bench 3.0 | 26% Terminal-Bench v3.0 | 26.5% Leaderboard, high, Grok Build | −0.5 pts |
| DeepSWE | 67% DeepSWE v1.1, Pass@1, xhigh, mini-swe-agent, run by Datacurve (2.4). High = 65.9% | 67.5% Leaderboard, medium, mini-swe-agent | −0.5 pts |
| Harvey LAB (Legal Agent Benchmark) | 15.8% Harvey LAB (Vals) | 15.8% Leaderboard | Same |
| CyberGym | 79.7% CyberGym, mean reproduced, high, unrestricted, Grok Build harness (7.1) | 79.7% 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. 34 of them aren't in the launch post.
Model Card: Grok 4.6 Aug 12, 2026
Capabilities
- CursorBench 3.2 70.8
- APEX-SWE 56.4
- FrontierCode 1.1 61.3
- DeepSWE 67
- SWE-Marathon 31.9
- Terminal-Bench 3.0 26
- GDPval-AA v2 1753
- AA-Briefcase 1577
- APEX-Agents 57.5
- OfficeQA 63.2
- Harvey LAB (Legal Agent Benchmark) 15.8
- EEBench 60
- 3DCodeBench 54
- PartBench 58.5
- CADGenBench 40.9
- Parametric CAD Bench
- DeepSearchQA 81.6
Hallucination
Safety
- General refusals (xAI) 0.93
- Child safety (xAI) 0
- Autointent-Bio 100
- Autointent-Chem 100
- FORTRESS-RN 97.9
- Self-harm refusals (xAI) 0.84
Robustness
Alignment
- MASK
- xAI sycophancy eval 0.04
Bio and chem risk
- Virology Capabilities Test (VCT) 67.4
- Biosecurity VCT 47.8
- BioUseBench 90.7
- WMDP
- LAB-Bench 80.7
- ProtocolQA Open-Ended (xAI) 79.6
- BixBench 93.8
Cyber risk
- CyberGym 79.7
- CVE-Bench 39.8
- SecureCodeReview (xAI) 58.7
- HackerBench 6.9
AI R&D risk
- SpaceXAI MTS Eval 61.1
- InferenceEval 46.9
- KernelBenchInternal 37.2