Llama 4 Scout Instruct: benchmark results
Meta's smaller natively multimodal Llama 4 MoE (109B total/17B active, 16 experts) with a 10M-token context (April 2025). Provider: Meta. Released 2025-04-05. Access: Open.
Aggregate Rating 1571 ± 6, rank #539 of 1656 rated models, from 1518 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BehaviorBench (Behavioral Traces) - Trade Amount Error (ProfileGen) | 10.3 | Median absolute error of the predicted trade amount (paper u | 100 |
| GSM-Plus-BN - Adding Operation | 74.55 | Accuracy (%; adding operation perturbation, an extra arithme | 100 |
| GSM-Plus-BN - Integer-Decimal-Fraction Conversion (CoT) | 79.52 | Accuracy (%; integer-decimal-fraction conversion perturbatio | 100 |
| GSM-Plus-BN - Numerical Substitution | 85.23 | Accuracy (%; numerical substitution perturbation, numbers re | 100 |
| GSM-Plus-BN - Numerical Substitution (CoT) | 88.42 | Accuracy (%; numerical substitution perturbation, numbers re | 100 |
| GSM-Plus-BN - Reversing Operation (CoT) | 88 | Accuracy (%; reversing operation perturbation, the result gi | 100 |
| HELM Long Context - InfiniteBench En.Sum | 17.6 | ROUGE-L | 100 |
| HELM VHELM - Blink Perception - Relative Depth | 96.77 | Prefix Quasi-Exact Match (%) | 100 |
| HumanAgencyBench - Avoid Value Manipulation | 66.88 | HAB dimension score (0-100, o3-judged rubric) | 100 |
| Open PL LLM - PolEmo2-IN (generative, 5-shot) | 89.34 | Accuracy (%) | 100 |
| SEA-HELM (Tamil) - Summarization | 14.4 | Normalized Score | 100 |
| SEA-HELM (Thai) - Question Answering | 89.1 | Normalized Score | 100 |
Interactive version: theaggregate.ai/model?slug=llama-4-scout-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.