Llama 3.2 3B Instruct: benchmark results
Meta Llama 3.2 3B instruction-tuned checkpoint. Provider: Meta. Released 2024-09-25. Access: Open.
Aggregate Rating 1365 ± 5, rank #1352 of 1656 rated models, from 1542 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| DeEscalWild | 36.1 | Realism score (0-100) from a Gemini 3.1 Pro judge rating beh | 100 |
| DeEscalWild - ROUGE-L | 11.3 | ROUGE-L (0-100) of the generated civilian turns against the | 100 |
| DeEscalWild - Realism (GPT-5.4 Judge) | 35.5 | Realism score (0-100) from a GPT-5.4 judge rating behavioura | 100 |
| StakeBench - Revealed Side | 53.4 | Strict accuracy (%) on G2: which side (YES or NO) a position | 100 |
| AGC-Bench - creatset | 1.63 | Dataset z-score | 98.8 |
| LA Leaderboard - Spanish Law Exams | 40.34 | Accuracy (%) | 93.4 |
| LA Leaderboard - Fake News Spanish | 80.77 | Accuracy (%) | 91.2 |
| LA Leaderboard - TELEIA | 57.14 | Accuracy (%) | 91.2 |
| SEA-HELM (Thai) - Summarization | 24.2 | Normalized Score | 91.2 |
| SLM Trustworthiness - Machine Ethics | 82.88 | TrustLLM machine-ethics accuracy (%; sub-task mean) | 90 |
| SLM Trustworthiness - Privacy | 60.15 | TrustLLM privacy accuracy (%; sub-task mean) | 90 |
| IndicIFEval-Ground - Bengali | 71.4 | Prompt-level loose accuracy (%, 0-100) on the Bengali prompt | 88.9 |
Interactive version: theaggregate.ai/model?slug=llama-3-2-3b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.