Llama 4 Scout: benchmark results
Meta's open Llama 4 Scout, a 109B sparse MoE (17B active) long-context multimodal model (April 2025). Provider: Meta. Released 2025-04-05. Access: Open.
Aggregate Rating 1535 ± 5, rank #656 of 1656 rated models, from 760 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CARE-Bench | 50.4 | Macro-F1 (%, unprompted) | 100 |
| Ebisu - JF-ICR | 60.64 | Accuracy (%) | 100 |
| MacroLens - Real-Estate Sale Price | 154.8 | Mean absolute percentage error (%; sale price predicted from | 100 |
| Nejumi 4 - BFCL - Irrelevance Detection | 100 | Accuracy (%) | 99.7 |
| AGC-Bench - sdat | 0.67 | Dataset z-score | 98.8 |
| GTA-Workflow - Perception | 15.79 | Root success rate (%) on the tasks that require perception t | 95.8 |
| AGC-Bench - unfun_corpus | 0.73 | Dataset z-score | 93.6 |
| GTA-Workflow - Leaf Success Rate | 27.51 | Leaf success rate (%): share of leaf checkpoints scoring abo | 91.7 |
| GTA-Workflow - Root Score | 3.65 | Mean root checkpoint score (0-10) on GTA-Workflow, the open- | 91.7 |
| Phare - Bias Resistance | 67.1 | Score (%) | 90.9 |
| Ebisu - Overall | 37.78 | Mean Task Score (%) | 90.5 |
| AMVICC - MMVP Individual Accuracy - Presence of Specific Features | 95.71 | Individual accuracy (%; the 70 MMVP questions on 35 image pa | 90 |
Interactive version: theaggregate.ai/model?slug=llama-4-scout · How It Works · Data refreshed daily, snapshot 2026-10-20.