Llama 4 Scout
Llama 4 Scout is the smaller April 2025 Llama 4 model: a 16-expert mixture-of-experts with 17B active parameters out of 109B. Meta rates it for a 10M-token context window. It accepts text and images and is widely hosted at low per-token prices.
Specifications
- Context window
- 10Mabout 7,864,000 words, or 15,730 pages
- Max output
- Not published
- Input price
- $0.10per 1M tokens
- Output price
- $0.30per 1M tokens
- Cached input
- Not published
- Weights
- OpenLlama 4 Community License
- Parameters
- 109B total, 17B activeMoE
- Input types
- Text, Images
- Output types
- Text
- Reasoning mode
- No
- Tool calling
- Yes
- Released
- Apr 5, 2025Legacy
DeepInfra serverless price for Llama-4-Scout-17B-16E (DeepInfra serves it with a 320K context, well below the official 10M).
What it costs in practice
| Workload | Monthly cost at list price |
|---|---|
| Support chatbot30,000 conversations a month, 2,000 input and 400 output tokens each | $9.60 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $2.88 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $7.80 |
Standard rates, no caching or batch discounts. Run your own numbers.
Strengths
- Officially rated for a 10M-token context window
- Runs on a single H100 with on-the-fly int4 quantization
- Low hosted prices
Trade-offs to weigh
- Hosts usually serve far less than the 10M context
- Llama 4 Community License rather than an OSI license
Where it fits
Meta positions Llama 4 Scout for:
Where you can use it
- Hugging Face
- Ollama
- DeepInfra
Run it yourself
Released under Llama 4 Community License. Approximate memory needed just to load the weights:
| Precision | Memory for weights |
|---|---|
| 16-bit (BF16/FP16) | 218 GB |
| 8-bit (FP8/INT8) | 109 GB |
| 4-bit (Q4_K_M, about 4.85 bits) | 66 GB |
This is a mixture-of-experts model: all 109B total parameters must sit in memory, but only about 17B are used per token, so it runs faster than a dense model of the same size.
Add memory for the KV cache (it grows with context length and concurrent users) and 1-2 GB of runtime overhead. Size it precisely in the VRAM calculator.
Alternatives to consider
| Model | Tier | Context | Input / output per 1M | |
|---|---|---|---|---|
| Ministral 3 14BMistral AI | Efficient | 256K | $0.20 / $0.20 | Compare |
| GLM-5.3-FlashZ.ai (Zhipu AI) | Efficient | 1M | $0.15 / $0.50 | Compare |
| Gemma 4 31BGoogle | Efficient | 256K | $0.20 / $0.40 | Compare |
| Mistral Small 4Mistral AI | Efficient | 256K | $0.15 / $0.60 | Compare |
Frequently asked questions
How much does Llama 4 Scout cost?
Llama 4 Scout lists at $0.10 per million input tokens and $0.30 per million output tokens (checked October 10, 2026). DeepInfra serverless price for Llama-4-Scout-17B-16E (DeepInfra serves it with a 320K context, well below the official 10M). Confirm current rates on the provider's pricing page before you commit.
What is the context window of Llama 4 Scout?
Llama 4 Scout accepts up to 10,485,760 tokens of context (about 7,864,000 words, or 15,730 pages). Long prompts cost more and can be slower, so most apps send only what the task needs.
Is Llama 4 Scout open source?
Llama 4 Scout is an open-weight model released under the Llama 4 Community License license. You can download the weights from Hugging Face (meta-llama/Llama-4-Scout-17B-16E-Instruct) and run or fine-tune them yourself, within the license terms. Open weights isn't the same as fully open source: training data and code may not be published.
Can I run Llama 4 Scout locally?
Yes, if you have enough memory. The weights alone need about 66 GB at 4-bit quantization and about 218 GB at 16-bit, plus room for the KV cache. Use the VRAM calculator on LLModel.com to check a specific GPU or Mac.
What is Llama 4 Scout best for?
Meta positions Llama 4 Scout for long documents, multimodal, self-hosting, high volume. It accepts text, images as input, doesn't offer a separate reasoning mode and supports tool calling. Test it on a sample of your own tasks before you choose it for production.
Sources
- huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct
- deepinfra.com/pricing
- huggingface.co/unsloth/Llama-4-Scout-17B-16E-Instruct/raw/main/config.json
- ollama.com/library/llama4
Facts checked against these pages on October 10, 2026. Spot an error? Tell us.
Still choosing a model?
Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.