llama4-scout-maverick-specific-parameters

IN premise — summaries/2026/08/24/wiki-LLaMA-chunk-1.md

Created 2026-08-24T17:11:14+00:00

Llama 4 Scout has 17B active parameters, 16 experts, 109B total parameters, and 10M context window; Llama 4 Maverick has 17B active parameters, 128 experts, 400B total parameters, and 1M context window.

Summary

These specs define two models that trade off in opposite directions: Scout keeps the same 17B active compute per token but uses fewer experts (16) and a much longer context window (10M tokens), making it suited for cheap, long-document workloads, while Maverick spends its extra 400B total parameters on 8x more experts (128) at the cost of a shorter 1M context, prioritizing per-token quality over throughput. For the system, this means routing decisions between them should weigh whether the task needs long context and speed (Scout) or maximum reasoning depth per token (Maverick).