- MMMU, 0-shot, accuracy 69.4
- MMMU Pro (avg of Standard and Vision) 52.2
- MathVista 70.7
- ChartQA, relaxed_accuracy 88.8
- DocVQA (test), anls 94.4
- LiveCodeBench (10/01/2024-02/01/2025), pass@1 32.8
- MMLU Pro, 0-shot, macro_avg/acc 74.3
- GPQA Diamond 57.2
- MGSM, average/em 90.6
- Best of MTOB (half book) eng->kgv/kgv->eng = 42.2/36.6 / MTOB (full book) = 39.7/36.3 42.2
- MMLU (5-shot) [pretrained] 79.6
- MATH (4-shot, em_maj1@1) [pretrained] 50.3
- MBPP (3-shot, pass@1) [pretrained] 67.8
- TydiQA (1-shot, average/f1) [pretrained] 31.5
Llama 4 Scout
Llama 4 Scout.
Meta Released 5 Apr 2025 72
Benchmarks with a score
0
Ranked first, out of five or more models
0
In the top three
Meta Apr 2025 Llama 4 Scout
Open weights
129.6
#162 of 271
Epoch Capabilities Index
- MMMU-Pro 52.2% #63 /72
- ZeroBench 3.00% #48 /66
- CritPt 0% #131 /160
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 | |
|---|---|---|---|---|
| MGSM | 90.6% | #10 of 52 | Model card | |
| Vectara Hallucination Leaderboard (2025 dataset) | 7.70% | #38 of 104 | Official leaderboard | |
| BullshitBench | 36.4% | #40 of 107 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 4.70% | #65 of 148 | Official leaderboard | |
| MMLU | 79.6% | #76 of 127 | Model card | |
| LMArena (Elo) | 1,321 | #207 of 333 | Official leaderboard | |
| Phare | 63.8% | #39 of 62 | Official leaderboard | |
| MMLU-Pro | 74.3% | #102 of 151 | Model card | |
| BullshitBench V2 | 20% | #85 of 122 | Official leaderboard | |
| SEA-HELM | 63.4 | #45 of 62 | Official leaderboard | |
| HELM Arabic | 75.6 | #21 of 28 | Official leaderboard | |
| HELM Arabic Enterprise | 62.1 | #18 of 23 | Official leaderboard | |
| WritingBench | 52.0 | #38 of 45 | Official leaderboard | |
| Fiction.LiveBench | 36% | #53 of 57 | Epoch AI | |
| Creative Writing v3 | 46.6 | #113 of 120 | Official leaderboard | |
| EQ-Bench 3 | 670 | #74 of 78 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 31.1 | #113 of 119 | Official leaderboard | |
| MTOB | 42.2 | #3 of 3 | Model card | |
| TydiQA | 31.5% | #2 of 2 | Model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ForecastBench | 59.0 | #22 of 86 | Official leaderboard | |
| MATH Level 5 | 62.3% | #42 of 94 | Epoch AI | |
| Misguided Attention | 35.9% | #29 of 45 | Official leaderboard | |
| GPQA Diamond | 51.8% | #142 of 219 | Epoch AI | |
| MATH | 50.3% | #63 of 93 | Model card | |
| OTIS Mock AIME 2024-2025 | 7.78% | #158 of 202 | Epoch AI | |
| HiPhO | 29.9% | #28 of 35 | Official leaderboard | |
| PolyMath | 20.9% | #31 of 38 | Official leaderboard | |
| CritPt | 0% | #131 of 160 | Epoch AI | |
| DTBench | 29.8% | #146 of 178 | Official leaderboard | |
| ARC-AGI-2 | 0% | #90 of 95 | Official leaderboard | |
| ARC-AGI | 0.50% | #91 of 92 | Official leaderboard | |
| FrontierMath | 0% | #70 of 70 | Epoch AI |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MBPP | 67.8% | #43 of 66 | Model card | |
| BigCodeBench-Hard | 16.9% | #68 of 96 | Official leaderboard | |
| LiveSQLBench | 18.9% | #28 of 36 | Official leaderboard | |
| LiveCodeBench | 32.8% | #59 of 73 | Model card | |
| LiveCodeBench v5 | 32.8% | #40 of 44 | Model card | |
| BigO(Bench) | 23% | #17 of 18 | Official leaderboard | |
| SWE-bench Verified | 9.06% | #57 of 60 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnakeBench | 21 | #118 of 183 | Official leaderboard | |
| BFCL V4 | 28.1% | #43 of 65 | Official leaderboard | |
| LisanBench | 126.3 | #109 of 133 | Official leaderboard | |
| Berkeley Function Calling Leaderboard (BFCL) | 45.4% | #57 of 67 | Official leaderboard | |
| LLM Chess | -402 | #108 of 124 | Official leaderboard | |
| PinchBench 2.0 | 3.19% | #57 of 57 | Official leaderboard | |
| Galileo Agent Leaderboard | 0.63 | #31 of 31 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LegalBench | 72% | #109 of 138 | Official leaderboard | |
| MortgageTax | 57.8% | #68 of 86 | Official leaderboard | |
| CorpFin v2 | 46.8% | #100 of 119 | Official leaderboard | |
| TaxEval v2 | 55.2% | #123 of 129 | Official leaderboard | |
| MedCode | 23.3% | #93 of 96 | Official leaderboard | |
| ProfBench | 31.2% | #71 of 72 | Official leaderboard | |
| MedScribe | 50.6% | #97 of 98 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU | 69.4% | #35 of 136 | Official leaderboard | |
| ChartQA | 88.8% | #11 of 37 | Model card | |
| SpatialViz-Bench | 34.2% | #4 of 8 | Epoch AI | |
| MathVista | 70.7% | #41 of 64 | Model card | |
| ZeroBench | 3.00% | #48 of 66 | Official leaderboard | |
| VHELM | 38.0 | #30 of 41 | Official leaderboard | |
| Vision Arena | 1,128 | #107 of 136 | Official leaderboard | |
| MMMU-Pro | 52.2% | #63 of 72 | Model card | |
| RoboRewardBench | 1.49 | #18 of 19 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SimpleSafetyTests | 97% | #56 of 75 | Official leaderboard | |
| HELM Safety | 87.3 | #57 of 75 | Official leaderboard | |
| AIR-Bench 2024 | 52.3% | #64 of 82 | Official leaderboard | |
| Anthropic Red Team | 96.5% | #64 of 75 | Official leaderboard | |
| KORA | 17.7% | #37 of 42 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Capabilities | 64.4 | #33 of 60 | Official leaderboard | |
| ACCoRD | 49.7 | #112 of 150 | Official leaderboard | |
| SAGE (Vals) | 34.8% | #69 of 81 | Official leaderboard | |
| Conceptual Reasoning Index | 23.1 | #130 of 149 | Official leaderboard | |
| LMCA | 12% | #123 of 139 | Epoch AI |
Meta's models over time
Meta's models in color, against every other model on the Epoch Capabilities Index, by release date.
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. 1 of them isn't in the launch post.