Meta AI·released 2023-07-181 source
Llama 2-70B benchmark scores: 14 benchmarks tracked, leading 1. 21% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 1 of 14 benchmarks·best result 87.6 on TriviaQA (unverified)·3 of 14 independently reproduced
Head-to-headLlama 2-70B vs Claude 2Leads on: TriviaQA
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| TriviaQA | 87.6accuracy (%) | unverified· optimizedT1 | 2023-07-18Llama 2: Open Foundation and Fine-Tuned Chat Models |
| HellaSwag | 80.4accuracy (%) | self-reported· optimizedT1 | 2023-07-18Qwen Technical Report |
| LAMBADA | 78.9accuracy (%) | self-reported· optimizedT1 | 2023-07-18Qwen Technical Report |
| ARC AI2 | 71.07accuracy (%) | unverified· optimizedT1 | 2023-07-18The Falcon Series of Open Language Models |
| GSM8K | 69.6accuracy (%) | unverified· optimizedT1 | 2023-07-18Stanford HELM |
| PIQA | 65.6accuracy (%) | self-reported· optimizedT1 | 2023-07-18Qwen Technical Report |
| Winogrande | 60.4accuracy (%) | unverified· optimizedT1 | 2023-07-18Llama 2: Open Foundation and Fine-Tuned Chat Models |
| MMLU | 59.87accuracy (%) | unverified· optimizedT1 | 2023-07-18Stanford CRFM Leaderboard |
| BBH | 53.2accuracy (%) | self-reported· optimizedT1 | 2023-07-18Qwen Technical Report |
| OpenBookQA | 46.93accuracy (%) | unverified· optimizedT1 | 2023-07-18Llama 2: Open Foundation and Fine-Tuned Chat Models |
| MATH level 5 | 3.29accuracy (%) | reproduced· optimizedT1 | 2023-07-18Epoch AI |
| DTBench | 2.75accuracy (%) | unverifiedT1 | 2023-07-18https://conceptualreasoning.ai/dtbench |
| GPQA diamond | 1.77accuracy (%) | reproduced· optimizedT1 | 2023-07-18Epoch AI |
| OTIS Mock AIME 2024-2025 | 0accuracy (%) | reproduced· optimizedT1 | 2023-07-18Epoch AI |
Claims drawn from cited facts, not live model generation.
Developed by Meta AI, this model is a large-scale 70 billion parameter system released in 2023. At this scale, the trade-off is greater computational expense and more headroom for capability compared with smaller models, which favor efficiency and lower cost.
4 cited facts
This model is scored on 14 tracked benchmarks. Of those, it currently holds one top score. Its results have been independently reproduced, meaning outside evaluation backs the scores up and the record can be trusted as verified.
3 cited facts
On average, this model trails the SOTA leader by 40.47 points, a score gap under a disclosed harness that reads as a wide gap—well behind the leaders. It holds the top score on 1 benchmark, indicating genuine but limited category leadership.
2 cited facts
2 facts cross-checked across data sources: 2 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: curated_param_counts, epoch_benchmarks
Llama 2-70B is an AI model developed by Meta AI, released 2023-07-18. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Llama 2-70B has recorded scores on 14 benchmarks, each shown with its evidence status.
Llama 2-70B has recorded scores on 14 benchmarks — PIQA, OpenBookQA, Winogrande, GPQA diamond, MATH level 5, OTIS Mock AIME 2024-2025, and 8 more. The full table above shows each score with its evidence status.
3 of 14 recorded scores (21%) are independently reproduced rather than self-reported by the lab.
Llama 2-70B has 14 tracked claims: 3 independently reproduced, 4 self-reported, 7 unverified.