Llama 3.1-70B places in the top three on 6 benchmarks.
Meta Jul 2024 Open 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 | |
|---|---|---|---|---|
| TriviaQA | 89.8% | #3 of 40 | Lab model card | |
| MuSR | 45.7% | #9 of 98 | Official leaderboard | |
| MMLU-CF | 70.4% | #4 of 37 | Official leaderboard | |
| ARC-Challenge | 94.8% | #9 of 49 | Lab model card | |
| AlpacaEval 2.0 | 38.1% | #9 of 47 | Official leaderboard | |
| CulturalBench | 54.2% | #3 of 15 | Official leaderboard | |
| Open Arabic LLM Leaderboard v1 | 52.5% | #12 of 55 | Official leaderboard | |
| HELMET | 41.7% | #4 of 16 | Official leaderboard | |
| SQuAD v2 | 81.8% | #2 of 8 | Lab model card | |
| WinoGrande | 83.5% | #12 of 47 | Lab model card | |
| CommonSenseQA | 84.1% | #5 of 19 | Lab model card | |
| ThaiExam | 62.3% | #7 of 24 | Official leaderboard | |
| InfiniteBench/En.QA | 36.7 | #1 of 3 | Lab model card | |
| PIQA | 83.8% | #13 of 37 | Lab model card | |
| MLogiQA | 52.4% | #6 of 17 | Official leaderboard | |
| MMLU | 86% | #49 of 131 | Lab model card | |
| RewardBench | 84% | #20 of 48 | Official leaderboard | |
| MGSM | 86.9% | #24 of 57 | Lab model card | |
| IFEval | 87.5% | #59 of 135 | Lab model card | |
| AGIEval | 64.6% | #18 of 40 | Lab model card | |
| BIG-Bench Hard | 81.6% | #34 of 73 | Lab model card | |
| Chinese SimpleQA | 40.2% | #13 of 27 | Official leaderboard | |
| AraGen | 50% | #21 of 42 | Official leaderboard | |
| SIQA | 52.2% | #12 of 24 | Lab model card | |
| ToRR | 44.4 | #8 of 14 | Official leaderboard | |
| MultiLoKo | 24.3% | #4 of 7 | Official leaderboard | |
| SQuAD | 81.8% | #3 of 5 | Lab model card | |
| Seal-0 | 0% | #7 of 11 | Official leaderboard | |
| BoolQ | 79.4% | #18 of 28 | Lab model card | |
| Vectara Hallucination Leaderboard (HHEM) | 5.00% | #97 of 148 | Official leaderboard | |
| AgentIF | 56.3% | #8 of 12 | Official leaderboard | |
| GRE Verbal | 166.0 | #4 of 6 | Lab model card | |
| AP Physics | 78.6% | #2 of 3 | Lab model card | |
| GMAT Verbal | 85.1% | #2 of 3 | Lab model card | |
| InfiniteBench/En.MC | 78.2% | #2 of 3 | Lab model card | |
| Multilingual MMLU (Meta) | 78.2% | #2 of 3 | Lab model card | |
| SAT Reading | 71.4% | #2 of 3 | Lab model card | |
| ZeroSCROLLS/QuALITY | 90.5% | #2 of 3 | Lab model card | |
| LMArena (Elo) | 1,293 | #233 of 335 | Official leaderboard | |
| SuperGPQA | 34.9% | #42 of 59 | Official leaderboard | |
| QuALITY | 82.8% | #5 of 7 | Lab model card | |
| RACE | 59% | #6 of 8 | Lab model card | |
| WorldSense | 61.1% | #3 of 4 | Lab model card | |
| LongBench v2 | 36.2% | #25 of 33 | Official leaderboard | |
| Needle In A Haystack | 97.5% | #7 of 9 | Lab model card | |
| MMLU-Pro | 66.4% | #133 of 167 | Lab model card | |
| OpenBookQA | 47.6% | #21 of 24 | Lab model card | |
| Creative Writing v3 | 50.8 | #107 of 120 | Official leaderboard | |
| AP Calculus | 54.3% | #3 of 3 | Lab model card | |
| PenguinScrolls | 66.5% | #2 of 2 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GSM8K | 95.1% | #12 of 92 | Lab model card | |
| KOR-Bench | 50% | #11 of 33 | Official leaderboard | |
| ZebraLogic | 24.9% | #18 of 46 | Official leaderboard | |
| MATH | 68% | #51 of 99 | Lab model card | |
| GRE Quantitative | 158.0 | #4 of 6 | Lab model card | |
| GMAT Quantitative | 84% | #2 of 3 | Lab model card | |
| GRE Physics | 74.7% | #2 of 3 | Lab model card | |
| GPQA Main | 42.6% | #42 of 60 | Official leaderboard | |
| MATH Level 5 | 36.7% | #66 of 94 | Epoch AI run | |
| GPQA Diamond | 44.2% | #178 of 229 | Epoch AI run | |
| Omni-MATH | 24.2% | #7 of 9 | Official leaderboard | |
| DTBench | 60% | #144 of 180 | Official leaderboard | |
| A^3-Bench | 44.8% | #6 of 7 | Official leaderboard | |
| U-MATH | 34.3% | #26 of 30 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 3.61% | #188 of 212 | Epoch AI run |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MBPP | 86% | #8 of 71 | Lab model card | |
| BigCodeBench | 50.5% | #21 of 69 | Official leaderboard | |
| BigCodeBench-Hard | 25.4% | #44 of 96 | Official leaderboard | |
| MHPP | 33.8% | #11 of 23 | Official leaderboard | |
| HumanEval | 80.5% | #54 of 101 | Lab model card | |
| BigO(Bench) | 33.8% | #11 of 18 | Official leaderboard | |
| HumanEval+ | 74.4% | #22 of 27 | Lab model card | |
| SciCode | 0% | #18 of 18 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) | 84.8% | #2 of 41 | Lab model card | |
| When2Call | 37.8% | #3 of 10 | Official leaderboard | |
| BALROG | 27.9% | #17 of 33 | Official leaderboard | |
| CRMArena | 41.1% | #5 of 9 | Official leaderboard | |
| AceBench | 60.4% | #10 of 16 | Official leaderboard | |
| API-Bank | 90% | #2 of 3 | Lab model card | |
| Gorilla APIBench | 29.7% | #2 of 3 | Lab model card | |
| Nexus | 56.7% | #2 of 3 | Lab model card | |
| LisanBench | 151.3 | #99 of 133 | Official leaderboard | |
| MCP-Bench | 0.51 | #12 of 15 | Official leaderboard | |
| ToolComp | 37.2% | #28 of 32 | Official leaderboard | |
| ComplexFuncBench | 2.70% | #10 of 11 | Official leaderboard | |
| PinchBench 2.0 | 10.7% | #56 of 57 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedQA | 76.9% | #15 of 71 | Official leaderboard | |
| MultiFinBen | 14.1 | #5 of 9 | Official leaderboard | |
| TheAgentCompany | 1.70% | #6 of 7 | Official leaderboard | |
| TaxEval v2 | 56.2% | #122 of 129 | Official leaderboard | |
| CorpFin v2 | 38.9% | #113 of 119 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SGP-Bench | 57.4% | #7 of 22 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CyberSecEval Code Interpreter Abuse | 3.80% | #1 of 2 | Lab model card | |
| AgentHarm | 14% | #5 of 8 | Official leaderboard | |
| FORTRESS | 44.2% | #53 of 73 | Official leaderboard | |
| nocot-bench (NCRI) | 57.6 | #147 of 185 | Official leaderboard | |
| IHEval | 17.1% | #8 of 10 | Official leaderboard | |
| HELM Safety | 84.5 | #67 of 75 | Official leaderboard | |
| SimpleSafetyTests | 92.5% | #67 of 75 | Official leaderboard | |
| AIR-Bench 2024 | 42.5% | #74 of 82 | Official leaderboard | |
| Anthropic Red Team | 93.2% | #71 of 75 | Official leaderboard | |
| CyberSecEval Prompt Injection | 26% | #3 of 3 | Lab model card | |
| CyberSecEval Spear Phishing | 24% | #2 of 2 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Lite | 80.8 | #14 of 71 | Official leaderboard | |
| Political Science LLM Benchmark | 67.5% | #14 of 42 | Official leaderboard | |
| Open LLM Leaderboard v2 | 26.2 | #38 of 99 | Official leaderboard | |
| HELM Capabilities | 57.4 | #47 of 60 | Official leaderboard | |
| ACCoRD | 43.2 | #127 of 152 | Official leaderboard | |
| Conceptual Reasoning Index | 24.2 | #128 of 151 | Official leaderboard | |
| LMCA | 14.8% | #125 of 147 | Official leaderboard |
The Self-Reported Index fits one score to the 24 capability results Meta 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 119.2 to 125.3. 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 |
|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) | 84.8% BFCL, 0-shot, acc, instruction tuned (Instruction tuned models table) | 53.6% Leaderboard, prompted | +31 pts |
| IFEval | 87.5% IFEval, instruction tuned (Instruction tuned models table) | 82.1% Leaderboard, strict, share of instructions followed | +5.4 pts |
| MMLU-Pro | 66.4% MMLU-Pro (CoT), 5-shot, macro_avg/acc, instruction tuned (Instruction tuned models table) | 62.8% Leaderboard | +3.6 pts |
| GPQA Diamond | 46.7% GPQA, 0-shot, em, subset not stated, instruction tuned (Instruction tuned models table) | 44.2% Epoch AI run | +2.5 pts |
| GMAT Quantitative | 84% GMAT Quant., accuracy (Table 17) | 84% Leaderboard | Same |
Meta'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)
Every result in the longer documents the lab published with it. 25 of them aren't in the launch post.