llama.cpp
The C++ inference engine powering most local LLMs.
OPEN SOURCECPU-CAPABLE
Meta
Llama 4 Scout
10M-token context — the longest window of any model, open weights.
Open weightstextimageREL 2025-04-05
Compare this model →Verified 2026-06-12
Overview
Llama 4 Scout's headline feature is a 10M-token context window — by far the largest available — in an open-weights model that fits (quantised) on a single high-end GPU. $0.10/$0.30 per 1M via API.
Benchmark scores
Methodology →Intelligence14
Codingn/a
Reasoningn/a
Mathn/a
Sources and verification
- Intelligence: Artificial Analysis Intelligence Index · checked 2026-06-12
- Catalogue and pricing: OpenRouter model record · checked 2026-06-12
Strengths and weaknesses
Strengths
- 10M context — unmatched
- Open weights, single-GPU capable (quantised)
- Dirt cheap
Trade-offs
- Quality trails newer open models
- 16K max output
Reach for this model when
Whole-codebase or massive-corpus analysis where context length is everything.
Run something like it locally
Runs locally via llama.cpp / Ollama — quantised on 24GB+, full precision on multi-GPU.