Meta
Updated May 2026
Llama
The open-weights family that made local AI real
Meta (Facebook's parent company) has taken the opposite bet from OpenAI — release the model weights for anyone to download, run, and modify. This turned Llama into the backbone of the open-source AI ecosystem. Every Chinese lab, every startup fine-tuning their own model, and every researcher wanting to study how LLMs work builds on top of Llama. Llama 4 in 2025 made the jump to mixture-of-experts and native multimodality.
Variants
6
Best at
truly open weights with permissive license
Max context
10M
tokens
Pricing
Self-host
$0
per download from HuggingFace
Meta AI
$0
per free inside WhatsApp/Instagram
Via Together.ai
~$0.20 in / $0.60 out
per million tokens (Maverick)
Via Groq
~$0.11 in / $0.34 out
per million tokens (Scout)
Variants
Sort the table by context window or cost to find the right variant. Click any version below for a battle card with ranks, pricing notes, and official links.
| Modalities | ||||
|---|---|---|---|---|
Llama 4 Scout llama-4-scout | 10M | $0.50 / $1.50 | 2025 | textvision |
Llama 4 Maverick llama-4-maverick | 1M | varies / varies | 2025 | textvision |
Llama 4 Behemoth llama-4-behemoth | 1M | varies / varies | 2025 | textvision |
Llama 3.3 70B llama-3-3-70b | 128K | $0.60 / $0.80 | 2024 | text |
Llama 3.1 8B llama-3-1-8b | 128K | varies / varies | 2024 | text |
Code Llama code-llama | 100K | varies / varies | 2023 | code |
Battle card
Context rank
#1
within Llama
Capability rank
#2
modalities + reasoning
Weights
Open
self-hostable if licensed
Best fights to pick
- self-hosted long-context analysis
- running frontier-ish AI in a single GPU
- document retrieval
Rankings are Tendril directory ranks, computed from the model data shown here. Public benchmark leaderboards change often, so official docs and current benchmark pages should be checked before buying or deploying.
Learn
Lessons about this model
Structured lessons that cover Llama directly or put it in context alongside its rivals.
Check yourself
Quizzes
Short, mixed-difficulty quiz sets on Llama and its model family.
Open-Weight Families: Llama, Mistral, Qwen, DeepSeek, Gemma
7 questions
The open ecosystem that shook the industry.
Start quiz →Hands-on
Try these prompts
Ready-made prompts that show Llama at its best. Use them in your own AI workspace, then compare the output with what you learned in Tendril.
Local Llama via Ollama
BuildersPull a Llama model locally and ask it a simple task — feel the difference in latency and control.
ollama run llama4:scout >>> Summarize the French Revolution in five bullet points. Be specific about dates.
Private-data task that shouldn't leave your machine
CreatorsThe whole point of Llama is running locally on data you can't send to a third party.
You are a HIPAA-compliant local assistant. Here is a patient intake form. Extract symptoms, medications, and allergies into a structured summary for the doctor's review. Do NOT output anything beyond the structured summary.
Fine-tune persona for your domain
BuildersLlama's strength is that you can continue training — try giving it a system prompt that mimics a fine-tune.
You are a customer support agent for a small outdoor gear retailer called BaseCamp. Brand voice: plainspoken, outdoorsy, never saccharine. If a customer is upset, acknowledge first, fix second. Now respond to: 'My tent arrived with a broken zipper and I leave for a trip in 3 days.'
Print & keep
Printable reference
One-page summaries and flowcharts — great for desks, classrooms, or study sessions.
Go deeper
Official resources
Straight from the lab — docs, API references, and the chat surfaces you can try today.
Strengths
- truly open weights with permissive license
- huge community fine-tune ecosystem
- 10M-token context on Scout
Limits
- Behemoth still not released
- benchmark performance behind GPT/Claude/Gemini
- license excludes 700M+ MAU companies