Llama 3.3 70B: benchmark results
Meta's open Llama 3.3 70B dense instruction model, delivering near-405B quality at 70B (December 2024). Provider: Meta. Released 2024-12-06. Access: Open.
Aggregate Rating 1561 ± 5, rank #579 of 1656 rated models, from 355 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BenSyc - Binary Detection | 61.8 | Binary sycophancy detection macro-F1 (%) (sycophantic versus | 100 |
| DNBench - BIRD - Single - Few-shot | 40.5 | DNB-Score (0-100) | 100 |
| DNBench - Complex - Zero-shot | 37.1 | DNB-Score (0-100) | 100 |
| DNBench - Component - Foreign Key Score | 17.4 | Component Score (0-100) | 100 |
| DNBench - Component - LLM Judge | 38.3 | Component Score (0-100) | 100 |
| DNBench - Component - Primary Key F1 | 63.7 | Component Score (0-100) | 100 |
| DNBench - Single - Few-shot | 45.1 | DNB-Score (0-100) | 100 |
| DNBench - Spider - Complex - Zero-shot | 38.8 | DNB-Score (0-100) | 100 |
| DNBench - Spider - Single - Few-shot | 47.1 | DNB-Score (0-100) | 100 |
| DiagFlowBench | 85 | Step Accuracy (self-reported) | 100 |
| DiagFlowBench - Correct Abstention | 81.3 | Correct abstention (%): share of injected off-procedure oper | 100 |
| FinEval 6.0 - Safety Awareness | 83.31 | Score (0-100) | 100 |
Interactive version: theaggregate.ai/model?slug=llama-3-3-70b · How It Works · Data refreshed daily, snapshot 2026-10-20.