Llama 4 Scout at a glance
Llama 4 Scout (109B total, 10M-token context) and Maverick (400B total) are natively multimodal MoE models with 17B active parameters, distilled from the still-training Behemoth.
What changed with Llama 4 Scout
On April 5, 2025, Meta released Llama 4 Scout and Llama 4 Maverick for download on llama.com and Hugging Face. They are Meta's first mixture-of-experts models and are natively multimodal. Both use 17 billion active parameters: Scout has 16 experts and 109 billion total parameters, while Maverick has 128 experts and 400 billion total.
Meta said Scout fits on a single NVIDIA H100 GPU and supports a 10 million token input context, and that Maverick beats GPT-4o and Gemini 2.0 Flash on many reported benchmarks. Both were distilled from Llama 4 Behemoth, a preview teacher model with 288 billion active and nearly two trillion total parameters that was still training. Meta AI in WhatsApp, Messenger and Instagram moved to Llama 4.
Llama 4 Scout in numbers
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Context windowas of · Source for Context window: ai.meta.com (opens in a new tab)
- 10M tokens
How it compares
Models that were current when this was published (released in the six months before it, newest version of each line), so these figures are from that time. Under each chart is the current best. (Models)
Current best (as of Nov 6, 2026): GPT-6 Cyber, 1.05M tokens
Where to compare Llama 4 Scout
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
General leaderboards
Why Llama 4 Scout matters for businesses
For businesses, Llama 4 widened the choice of self-hostable models with long context and image input. The Llama license carries its own conditions, and vendor benchmarks are only a starting point, so review the terms and test the models on your own tasks before committing.
Sources for this story
- The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation (opens in a new tab)Official
Meta AI · ai.meta.com
- meta-llama/Llama-4-Maverick-17B-128E-Instruct (opens in a new tab)Official
Hugging Face · huggingface.co
- Llama 4 Maverick model page (opens in a new tab)Benchmark
Artificial Analysis · artificialanalysis.ai
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #llm
- #meta
- #llama
- #open-weights
- #multimodal