Meta Llama 4 Scout (109B MoE)
Meta’s 109B Mixture-of-Experts model activating only 17B parameters per forward pass—delivering 70B-class intelligence at 17B inference speeds.
Provider
Meta AI (Llama 4 Community License)
Architecture
16-Expert MoE (109B Total / 17B Active) + iRoPE Long Context
Parameters
109B Total / 17B Active
Context Window
1,000,000 tokens (Max Output: 32,768 tokens)
API Pricing
Input: $0.18 / 1M · Cached: $0.09 / 1M · Output: $0.55 / 1M
VRAM (16K Ctx)
Q4_K_M: 68.48 GB · Q8_0: 118.73 GB · FP16: 222.2 GB
Hardware Target
NVIDIA H100 80GB, Dual RTX 5090 (64GB), or Apple M4 Max/Ultra (64GB+)
Interactive Token Cost Estimator (Llama 4 Scout (109B))
GPU Hardware Compatibility Matrix (Llama 4 Scout (109B) at 16K Context)
| GPU / Hardware Profile | VRAM | Q4_K_M (4.8-bit) | Q8_0 (8-bit) | Est. Speed (Q4) |
|---|---|---|---|---|
| RTX 5060 Ti 16GB 448 GB/s | 16 GB | REQUIRES MULTI-GPU (68.48 GB) | REQUIRES MULTI-GPU (118.73 GB) | ~6 tok/s |
| RTX 4090 24GB 1008 GB/s | 24 GB | REQUIRES MULTI-GPU (68.48 GB) | REQUIRES MULTI-GPU (118.73 GB) | ~13 tok/s |
| RTX 5090 32GB 1792 GB/s | 32 GB | REQUIRES MULTI-GPU (68.48 GB) | REQUIRES MULTI-GPU (118.73 GB) | ~24 tok/s |
| M4 Max 64GB 546 GB/s | 52 GB | REQUIRES MULTI-GPU (68.48 GB) | REQUIRES MULTI-GPU (118.73 GB) | ~7 tok/s |
| 2× 24GB (48GB) 920 GB/s | 48 GB | REQUIRES MULTI-GPU (68.48 GB) | REQUIRES MULTI-GPU (118.73 GB) | ~12 tok/s |
| H100 80GB 3350 GB/s | 80 GB | FITS IN VRAM (68.48 GB) | REQUIRES MULTI-GPU (118.73 GB) | ~211 tok/s |
| M4 Ultra 192GB 819 GB/s | 164 GB | FITS IN VRAM (68.48 GB) | FITS IN VRAM (118.73 GB) | ~52 tok/s |
| 8× H200 (1,128GB) 4800 GB/s | 1128 GB | FITS IN VRAM (68.48 GB) | FITS IN VRAM (118.73 GB) | ~302 tok/s |
Architectural Strengths & Target Workloads
- 17B active parameter footprint reaches 60–110+ tok/s on workstation GPUs and Apple Silicon
- Fits inside a single 80GB H100 or 64GB Dual-RTX 5090 / Apple M4 Max at 4-bit quantization
- Interleaved RoPE (iRoPE) supports up to 1M context tokens
Primary Engineering Workloads: Fast Local Agent Routing · Multimodal Code & UI Inspection · Single-Node H100 Serving.
Frequently Asked Questions (Llama 4 Scout (109B))
How much does Meta Llama 4 Scout (109B MoE) cost per 1M tokens?
Meta Llama 4 Scout (109B MoE) is priced at $0.18 per 1M uncached input tokens, $0.09 per 1M cached input tokens, and $0.55 per 1M output tokens (Groq / Together / Fireworks API or Local Workstation (Dual GPU / M4 Ultra)).
What is the context window of Meta Llama 4 Scout (109B MoE)?
It supports up to 1,000,000 tokens (1,000,000 tokens) and a maximum generation output of 32,768 tokens.