Llama 4 Scout

Llama 4 Scout

Quick answer: Llama 4 Scout is Meta's April 2025 open-weight multimodal MoE model featuring a 10M-token context window — the largest of any open-weight model at launch. It scores 90.7% on AI2 Reasoning Challenge and 74.3% on MMLU-Pro with 17B active parameters. Available under the Llama 4 License with open weights on Hugging Face.

At a Glance

Where Llama 4 Scout leads

• 10M-token context window — largest of any open-weight model at launch
• 90.7% AI2 Reasoning Challenge — strong multi-discipline scientific reasoning
• 74.3% MMLU-Pro — competitive academic knowledge
• Open weights — self-hostable
• Multimodal: text and image inputs
• MoE efficiency: 17B active params at inference

Where it lags

• 74.3% MMLU-Pro and 90.7% AI2 RC — below Llama 4 Maverick (80.5%, 93.8%)
• 16.9% BigCodeBench — limited coding performance
• Llama 4 License (not Apache 2.0) — commercial use restrictions apply

Best for: Open-weight deployments requiring extremely long context (book-length documents, whole-repository analysis); tasks where the 10M context window is the primary requirement.

What Llama 4 Scout Is

Llama 4 Scout is Meta's efficient model in the Llama 4 family, prioritising maximum context length over peak accuracy. Its 10M-token context window is 10× that of Llama 4 Maverick and GPT-4.1 — making it uniquely suited for tasks that require processing an entire codebase, book, or document collection in a single pass.

Like Maverick, Scout uses a Mixture of Experts (MoE) architecture with 17B active parameters, making inference efficient. However, Scout is designed with fewer experts (16 per token vs Maverick's 128) — enabling the extreme context extension at the cost of reduced peak accuracy.

At 90.7% AI2 Reasoning Challenge and 74.3% MMLU-Pro, Scout is a capable model across academic knowledge tasks, but Maverick outperforms it on standard benchmarks. For tasks that don't specifically require the 10M context window, Llama 4 Maverick is the recommended choice within the Llama 4 family.

Specifications

Pricing

Llama 4 Scout is available as open weights — free to download and self-host. Hosted API access is available via Meta AI and third-party providers (Together AI, Groq, etc.) at market rates.

Context Window

Llama 4 Scout has a 10,000,000-token context window — roughly 7,500 pages of text, or a large codebase in a single request. This is the largest context window of any open-weight model, and matches or exceeds all closed-source models except Gemini 1.5 Pro Extended.

Public Benchmark Scores

Llama 4 Scout vs Alternatives

Scout vs Maverick: 10× larger context window; lower accuracy on standard benchmarks. Choose Scout when the 10M context window is a hard requirement; choose Maverick for better performance on standard tasks.

Run Llama 4 Scout

# Via Together AI (hosted)
from together import Together

client = Together(api_key="YOUR_API_KEY")
response = client.chat.completions.create(
    model="meta-llama/Llama-4-Scout-17B-16E-Instruct",
    messages=[{"role": "user", "content": "Analyse this entire codebase..."}],
)
print(response.choices[0].message.content)

Related Models & Benchmarks

• Compare: Llama 4 Maverick · GPT-4.1 · Gemini 2.5 Flash
• Benchmarks: AI2 Reasoning Challenge · MMLU-Pro

Frequently Asked Questions

Specs from Meta's official Llama 4 announcement (April 5, 2025) and Benchgen evaluations. Last updated 2026-07-24.