Grok 4 Fast ranks first on 3 of the 96 benchmarks with five or more models scored.
Closed weights
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SimpleQA | 95% | #3 of 60 | Lab model card | |
| Fiction.LiveBench | 94.4% | #4 of 57 | Fiction.live leaderboard, via Epoch AI | |
| SnorkelSpatial | 84.8% | #2 of 27 | Official leaderboard | |
| LMArena (Elo) | 1,418 | #105 of 335 | Official leaderboard | |
| Phare | 66.5% | #31 of 62 | Official leaderboard | |
| FACTS Search | 67.3% | #17 of 31 | Official leaderboard | |
| Humanity's Last Exam | 20% | #104 of 154 | Lab model card | |
| SimpleQA Verified | 19.7% | #42 of 56 | Official leaderboard | |
| FACTS Parametric | 15.8% | #36 of 45 | Official leaderboard | |
| DesignArena ASCII Art | 1,103 | #65 of 77 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 19.7% | #99 of 104 | Official leaderboard | |
| xAI Factuality (Hallucination) eval | 12.1% | #2 of 2 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnorkelGraph | 75% | #3 of 24 | Official leaderboard | |
| A^3-Bench | 65.1% | #1 of 7 | Official leaderboard | |
| ForecastBench | 0.13 | #19 of 82 | Official leaderboard | |
| GPQA Diamond | 85.7% | #71 of 302 | Lab model card | |
| BRUMO 2025 | 95.8% | #11 of 42 | Official leaderboard | |
| SnorkelSequences | 70.2% | #7 of 25 | Official leaderboard | |
| AIME 2025 | 92% | #42 of 143 | Lab model card | |
| HMMT 2025 | 93.3% | #18 of 54 | Lab model card | |
| CMIMC 2025 | 85% | #14 of 33 | Official leaderboard | |
| DTBench | 82.7% | #79 of 180 | Official leaderboard | |
| HMMT Nov 2025 | 90.8% | #10 of 22 | Official leaderboard | |
| MathArena (Apex, IMO, USAMO) | 5.21% | #26 of 47 | Official leaderboard | |
| SMT 2025 | 84.4% | #23 of 40 | Official leaderboard | |
| ARC-AGI-2 | 5.28% | #67 of 98 | Official leaderboard | |
| ARC-AGI | 48.5% | #68 of 95 | Official leaderboard | |
| MathArena Apex Shortlist | 55.3% | #28 of 38 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| IOI (Vals) | 11.5% | #32 of 55 | Official leaderboard | |
| WeirdML | 42.9% | #67 of 112 | Official leaderboard | |
| DesignArena Website | 1,107 | #114 of 151 | Official leaderboard | |
| DesignArena Code | 1,091 | #114 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,063 | #112 of 139 | Official leaderboard | |
| DesignArena Game Dev | 1,056 | #116 of 143 | Official leaderboard | |
| DesignArena 3D Design | 1,039 | #113 of 136 | Official leaderboard | |
| DesignArena Data Visualization | 1,095 | #104 of 118 | Official leaderboard | |
| Vibe Code Bench v1 | 0% | #16 of 17 | Official leaderboard | |
| Vibe Code Bench | 0% | #117 of 120 | Official leaderboard | |
| DesignArena SVG | 966 | #101 of 103 | Official leaderboard | |
| Code Arena | 1,162 | #127 of 128 | Official leaderboard | |
| WebDev Arena | 1,149 | #105 of 105 | WebDev Arena Leaderboard, via Epoch AI |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnakeBench | 33 | #17 of 188 | Official leaderboard | |
| MCP-Universe | 32.5% | #7 of 30 | Official leaderboard | |
| LisanBench | 2749.6 | #37 of 133 | Official leaderboard | |
| LLM Chess | 735 | #35 of 123 | Official leaderboard | |
| OdysseyArena Repo System | 38.3% | #3 of 8 | Official leaderboard | |
| SnorkelWordle | 88% | #8 of 19 | Official leaderboard | |
| MCPMark | 24% | #13 of 28 | Official leaderboard | |
| Research-Eval | 66% | #1 of 2 | Lab model card | |
| X Browse (xAI) | 58% | #1 of 2 | Lab model card | |
| Agent Poker Bench | 1,034 | #10 of 17 | Official leaderboard | |
| OdysseyArena AI Trading | 5.70% | #5 of 8 | Official leaderboard | |
| OdysseyArena Turn on Lights | 14.2% | #5 of 8 | Official leaderboard | |
| CompileBench | 73.3% | #13 of 20 | Official leaderboard | |
| xbench-DeepSearch | 74% | #2 of 3 | Lab model card | |
| BrowseComp-ZH | 51.2% | #14 of 19 | Lab model card | |
| Toolathlon | 18.5% | #41 of 50 | Official leaderboard | |
| BrowseComp | 44.9% | #67 of 81 | Lab model card | |
| LMArena Search Arena | 1,162 | #31 of 37 | Official leaderboard | |
| τ³-bench Banking | 15.7% | #21 of 25 | Official leaderboard | |
| OdysseyArena Energy Dispatch | 0% | #7 of 8 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| TaxEval v2 | 75.7% | #13 of 129 | Official leaderboard | |
| CorpFin v2 | 66.9% | #16 of 119 | Official leaderboard | |
| SnorkelUnderwrite | 81.3% | #5 of 27 | Official leaderboard | |
| CaseLaw v2 | 65.7% | #10 of 53 | Official leaderboard | |
| Vals Finance Agent v1 | 37.1% | #20 of 47 | Official leaderboard | |
| ProfBench | 48.7% | #33 of 72 | Official leaderboard | |
| FinQA (Snorkel OpenEnv environment) | 73.5% | #11 of 22 | Official leaderboard | |
| SnorkelFinance | 73.5% | #11 of 22 | Official leaderboard | |
| MedScribe | 81.6% | #49 of 97 | Official leaderboard | |
| FailureSensorIQ | 64.1% | #14 of 24 | Official leaderboard | |
| LegalBench | 80.6% | #81 of 138 | Official leaderboard | |
| Finance Reasoning (Snorkel) | 40.5% | #14 of 21 | Official leaderboard | |
| Vals Finance Agent | 46.1% | #35 of 49 | Official leaderboard | |
| MedCode | 37.4% | #70 of 96 | Official leaderboard | |
| Vals Index v1 | 42.6% | #30 of 34 | Official leaderboard | |
| MortgageTax | 42.1% | #82 of 87 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| WorldVQA | 18.9% | #11 of 22 | Official leaderboard | |
| ClockBench | 3.30% | #37 of 37 | Official leaderboard | |
| Math Kangaroo 2025 (MathArena) | 66.8% | #22 of 22 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PCT (Polarized Contrastive Pairs) | 30.8% | #3 of 46 | Official leaderboard | |
| AgentDojo | 0% | #1 of 8 | Lab model card | |
| AgentHarm | 8.00% | #1 of 6 | Lab model card | |
| Lakera B3 (Backbone Breaker Benchmark) | 75.2% | #6 of 35 | Official leaderboard | |
| Cybench | 30% | #4 of 16 | Official leaderboard | |
| General refusals (xAI) | 0% | #3 of 8 | Lab model card | |
| Refusals + System Jailbreak (xAI) | 0% | #2 of 5 | Lab model card | |
| Virology Capabilities Test (VCT) | 54.5% | #13 of 31 | Lab model card | |
| BioLP-bench | 39% | #4 of 9 | Lab model card | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 50.9% | #40 of 71 | Official leaderboard | |
| Spiral-Bench v1.2 | 37.5 | #12 of 21 | Official leaderboard | |
| Refusals + User Jailbreak (xAI) | 0% | #3 of 5 | Lab model card | |
| Human Pathogen Capabilities Test | 46.8% | #56 of 86 | Official leaderboard | |
| Molecular Biology Capabilities Test | 44.3% | #63 of 90 | Official leaderboard | |
| SycophancyEval (answer sycophancy) | 10% | #3 of 4 | Lab model card | |
| MakeMeSay | 12% | #8 of 10 | Lab model card | |
| MASK | 53% | #16 of 19 | Lab model card | |
| Soft Bias (Internal) | 0.79 | #2 of 2 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CAIS Risk Index | 58.3 | #47 of 67 | Official leaderboard | |
| CAIS Text Capabilities Index | 19.0 | #52 of 68 | Official leaderboard | |
| CAIS Vision Capabilities Index | 37.7 | #47 of 53 | Official leaderboard | |
| SAGE (Vals) | 29.8% | #77 of 81 | Official leaderboard |
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 141.0 to 152.3. How we compute it.
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 |
|---|---|---|---|---|
| HMMT 2025 | 93.3% HMMT 2025 (no tools) pass@1 | 91.7% Leaderboard | +1.6 pts | |
| AIME 2025 | 92% AIME 2025 (no tools) pass@1 | 90.8% Leaderboard | +1.2 pts | |
| LMArena Search Arena | 1,163 LMArena Search Arena, grok-4-fast-search (menlo), #1 | 1,162 Leaderboard, search | +1 | The lab quotes the September 2025 Search Arena snapshot, while the board uses a later rating with style controls. |
| Cybench | 30% CyBench, unguided success rate 30.0 (Table 3) | 30% Leaderboard | Same |
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)
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 13 of them aren't in the launch post.