Llama 3.1 70B Instruct: benchmark results
Meta's mid-size open Llama 3.1 instruction-tuned model (70B dense) with a 128K context and multilingual support (July 2024). Provider: Meta. Released 2024-07-23. Access: Open.
Aggregate Rating 1565 ± 4, rank #561 of 1656 rated models, from 2281 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ADRD-Bench - Unified QA | 92.98 | Exact-match accuracy (0-1, shown times 100) on the 1,438 Alz | 100 |
| AbstentionBench - underspecified context - MMLU Math - F1 | 84.48 | Abstention F1 (%) | 100 |
| AbstentionBench - underspecified context - MMLU Math - Recall | 73.68 | Abstention Recall (%) | 100 |
| CRMArena - KQA (Act) | 47.9 | F1 (%) | 100 |
| CRMArena - PVI (Function Calling Prompt) | 37.4 | Exact match (%) | 100 |
| CRMArena - PVI (ReAct) | 37.7 | Exact match (%) | 100 |
| EPPCMinerBen (0-Shot) - Evidence Extraction | 78.87 | Micro-averaged F1 (%) of EPPC evidence extraction (the minim | 100 |
| EPPCMinerBen (1-Shot) - Code Classification | 64.93 | Micro-averaged F1 (%) of EPPC code classification (sentence- | 100 |
| EPPCMinerBen (1-Shot) - Evidence Extraction | 82.06 | Micro-averaged F1 (%) of EPPC evidence extraction (the minim | 100 |
| EPPCMinerBen (2-Shot) - Evidence Extraction | 82.84 | Micro-averaged F1 (%) of EPPC evidence extraction (the minim | 100 |
| European LLM Leaderboard - Accuracy | 70.57 | Average Accuracy (%) | 100 |
| FilBench - Classical NLP | 90.27 | Category Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=llama-3-1-70b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.