Llama 4 Maverick Instruct FP8: benchmark results
Llama 4 Maverick Instruct evaluated in FP8 quantization. Provider: Meta. Released 2025-04-05. Access: Open.
Aggregate Rating 1607 ± 8, rank #414 of 1656 rated models, from 926 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM SeaHELM - Flores (en-ta) | 51.98 | ChrF++ | 100 |
| HELM SeaHELM - Flores (th-en) | 59.27 | ChrF++ | 100 |
| HELM SeaHELM - XQuAD (Thai) | 77.14 | SQuAD macro-averaged F1 score | 100 |
| HELM VHELM - Mmmu - Sociology | 86.67 | Prefix Quasi-Exact Match (%) | 100 |
| Twinkle Eval - TW-Legal Benchmark v1 | 61.4 | Accuracy (%) | 100 |
| Nejumi 4 - BFCL - Relevance Detection | 100 | Accuracy (%) | 98.6 |
| HELM VHELM - Blink Perception | 84.43 | PEM | 97.9 |
| HELM VHELM - Mm Star Perception - Fine-grained Perception | 70.4 | Prefix Quasi-Exact Match (%) | 97.9 |
| FilBench - Cultural Knowledge | 76.75 | Category Score (%) | 97.3 |
| FilBench | 67.67 | FilBench Score (%) | 97.2 |
| TableVista - Partial | 52.9 | Accuracy (%) on the 3,000 TableVista questions under the Par | 96.4 |
| Twinkle Eval - TMMLU+ | 78.38 | Accuracy (%) | 96.4 |
Interactive version: theaggregate.ai/model?slug=llama-4-maverick-instruct-fp8 · How It Works · Data refreshed daily, snapshot 2026-10-20.