MetaactiveOpen Source

Llama 4 Scout

llama-4-scout

Llama 4 with 10M context window and 17B active params (16 experts).

Context Window

10M

tokens

Max Output

32.8K

tokens

Input Price

—

per 1M tokens

Output Price

—

per 1M tokens

Details

Familyllama-4

Parameters17Bx16E

Training Cutoff2025-03-01

ReleasedApril 5, 2025

Aliasesmeta-llama/Llama-4-Scout-17B-16E-Instruct

Capabilities

VisionFunctionsStreamingJSON ModeCodeTool UseMultimodal

Documentation

Evaluation Scores(4 benchmarks)

HumanEvalFunction-level Python code generation

72%

MMLU-ProHarder successor to MMLU

66.8%

MATH-500Competition-style math

61.5%

GPQA DiamondPhD-level science questions

43.1%

Quick Access

curl pikaainews.com/api/models/meta-llama-4-scout

npx pika-models info meta-llama-4-scout

Get API Access

Official

AWS Bedrock

Amazon Bedrock. Official partnerships with Anthropic, Meta, Mistral, Cohere.

Third-Party Providers & Aggregators

Cerebras

Wafer-scale inference. 1000+ tokens/sec for select models.

DeepInfra

Lowest per-token rates for open-source models.

Fireworks AI

Fastest inference engine. Multimodal support, HIPAA/SOC2.

Groq

Ultra-fast LPU inference. Best latency for real-time apps.

OpenRouter

500+ models, one API key. Pay-per-token, no minimums.

SiliconFlow

China-optimized inference. Strong Qwen/DeepSeek support.

Together AI

Fast open-source model inference. Sub-100ms latency.