Llama 3.1 405B Instruct FP8: benchmark results
Llama 3.1 405B Instruct evaluated in FP8 quantization. Provider: Meta. Released 2024-07-23. Access: Open.
Aggregate Rating 1617 ± 13, rank #389 of 1656 rated models, from 423 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - jaster (2-shot) - JCoLA (in-domain) | 85 | Exact match (%) | 100 |
| Open PL LLM - 8TAGS (generative, 5-shot) | 80.56 | Accuracy (%) | 100 |
| Open PL LLM - Belebele (multiple choice, 5-shot) | 93.44 | Accuracy (%) | 100 |
| OpenCompass Agent - T-Eval - Chinese (CompassBench 2409) | 87.6 | Score (%) | 100 |
| OpenCompass Knowledge - Science - Chinese (CompassBench 2409) | 72 | Score (%) | 100 |
| OpenCompass LLM - Agent (CompassBench 2409) | 87.7 | Score (%) | 100 |
| Open PL LLM - Generative | 72.51 | Average Generative Score (%) | 99.7 |
| Open PL LLM - KLEJ NER (generative, 5-shot) | 64.09 | Accuracy (%) | 99.7 |
| Open PL LLM - PPC (generative, 5-shot) | 82.5 | Accuracy (%) | 99.7 |
| Open PL LLM - PolQA Closed Book (generative, 5-shot) | 85.98 | Levenshtein Similarity (%) | 99.7 |
| Open PL LLM Leaderboard | 69.44 | Average Score (%) | 99.7 |
| EuroEval Swedish | 76.31 | Average Score (%) | 99.5 |
Interactive version: theaggregate.ai/model?slug=llama-3-1-405b-instruct-fp8 · How It Works · Data refreshed daily, snapshot 2026-10-20.