Llama

Meta, US

⭐⭐⭐⭐ 4.3/5

Editorial estimate compiled from public vendor data — not a user-review average. How we rate

Visit Website →

📊 Quick Facts

Pricing ModelOpen Source

Categorychatbots

Users500M+ downloads

Founded2023

HQMenlo Park, CA, USA

EmployeesPart of Meta (70,000+ total employees)

Monthly Visits500M+ MAU (Meta AI)

LanguagesEnglish, Spanish, French, German, Portuguese, Italian, Japanese, Korean, Chinese, Arabic, 30+ languages

PlatformWeb (meta.ai), WhatsApp, Instagram, Facebook, Messenger, Ray-Ban Meta, Hugging Face, Ollama, API, AWS, Azure, GCP

DeploymentCloud (Meta AI) + Open Source + Self-Hosted

Data Snapshot2026-09-18

📝 About Llama

Llama is Meta's open-weight LLM family. As of June 2026, Llama 4 Maverick (400B total/17B active MoE, 1M context) is the flagship, with Llama 4 Scout (109B/17B active, 10M context) for long-context work. Llama 4 Behemoth (~2T params) was previewed but never publicly released. All models use mixture-of-experts architecture with native multimodal (text+image) early fusion. Available under Llama 4 Community License (commercial use below 700M MAU).

💰 Pricing

Free: Open weights (free download)Model weights free under Llama 4 Community License
Hosted: Varies by providerHosted inference via Together, Groq, Fireworks, etc.
Enterprise: Custom licensingRequired for >700M MAU products

✨ Key Features

Llama 4 Maverick: 400B MoE flagship, 1M context, early-fusion multimodalLlama 4 Scout: 109B MoE, industry-leading 10M token context windowNative multimodal: text + image understanding via early fusioniRoPE architecture: interleaved NoPE/RoPE layers for ultra-long contextOpen weights under Llama 4 Community LicenseWidely supported by inference providers (Together, Groq, Fireworks, vLLM)Strong self-hosting ecosystem (vLLM, SGLang, Ollama)Behemoth (~2T params) paused May 2026; no Llama 5 announced

🏷️ Tags

open-sourcellmmetacoding

⚖️ Pros & Cons

✓ Pros

Completely free and open source
Commercial-friendly license
Capabilities close to GPT-4
Massive ecosystem and community
Self-hostable for complete privacy

✗ Cons

Technical setup required
Needs good hardware for best performance
Less polished than closed alternatives
No official consumer app
Requires third-party tools

🎯 Use Cases

• chatbots

• coding

• self-hosted

✔ Best For

Self-Hosted AIOpen Source DevelopmentResearchEnterprise AICode GenerationMultilingual Tasks

🧠 Models

Llama 4 Maverick
Llama 4 Scout

🔗 Integrations

WhatsAppInstagramFacebookMessengerRay-Ban MetaHugging FaceOllamaReplicateAWS BedrockAzure AIGoogle Cloud VertexLangChainLlamaIndex

📈 Benchmarks

MMLU: 88.6%

HumanEval: 91.2%

GSM8K: 94.5%

MATH: 79.8%

MMMU: 73.2%

10M token context window

🔀 Alternatives

DeepSeek Mistral

📅 History

Feb 2023 Llama 1 - Initial Llama release; research-only access

Jul 2023 Llama 2 - Commercial-friendly open weights; 7B, 13B, 70B models

Apr 2024 Llama 3 - Llama 3 8B and 70B; state-of-the-art open models

Jul 2024 Llama 3.1 405B - First frontier-level open model at 405B parameters

Apr 2025 Llama 4 Herd - Llama 4 Maverick (400B MoE) and Scout (109B MoE, 10M context) released

May 2026 Behemoth Paused - Llama 4 Behemoth (~2T params) public release paused amid capability concerns

❓ FAQ

Is Llama free?

Yes, Llama models are open source. Meta AI chatbot is free on WhatsApp, Instagram, and Facebook.

Can I use Llama commercially?

Yes, Llama is released under the Llama 3.1 Community License, allowing commercial use.

What is Llama 4?

Llama 4 (Scout & Maverick) was released in April 2025 with multimodal capabilities, improved reasoning, and 10M token context window.

🇨🇳 China Insight

Llama is widely used in China by developers and companies. It's often fine-tuned for Chinese language tasks alongside local models.

🔒 Data & Compliance

Data PolicyOpen source model; training data from public sources. Meta AI processes data per Meta privacy policy. EU availability limited.

ComplianceGDPR (limited in EU), CCPA, US AI Safety commitments

Content ModerationMeta AI has integrated safety features; open-source models include responsible use guidelines

Data RetentionOpen source: user-controlled. Meta AI: per Meta privacy policy.

🌐 Social & Community

GitHub X/Twitter

💬 Reviews

HuggingFace: 4.8/5,GitHub: 100K+ stars (combined),Open LLM Leaderboard: Top 5

ℹ️ How this page is built

This page is compiled by an automated pipeline from publicly available vendor information. It is not a hands-on review, and no editor reviews it before publication.

Ratings are editorial estimates derived from public vendor data (features, pricing, ecosystem coverage). They are not user-review averages, and we do not publish user ratings.

Data Snapshot shows when this entry was last refreshed from our directory dataset. Pricing and model line-ups change frequently, so confirm critical pricing on the vendor’s own site.

Read our directory methodology · Report an error