Meta Llama 4 family
https://ai.meta.com/blog/llama-4-multimodal-intelligence/
Meta's Llama 4 family (released 2025-04) — Scout (109B MoE, 17B active), Maverick (400B MoE, 17B active), Behemoth (2T MoE, 288B active; preview). First Llama generation to use mixture-of-experts and to be multimodal-native. Substrate for many OSS agents (Open-Interpreter, Ollama-served local agents, vLLM-served enterprise inference). Llama community license (commercial use permitted under 700M MAU threshold).
At a glance
- Type
- Open-weights frontier model family (Llama 4 Scout / Maverick / Behemoth)
- Tier
- T1
- Created
- 2023-02 (Llama 1); 2023-07 (Llama 2); 2024-04 (Llama 3); 2024-12 (Llama 3.3); 2025-04 (Llama 4)
- Latest release
- Llama 4 Scout + Maverick (2025-04); Behemoth in preview as of 2025
- License
- Llama 4 community license (Meta-custom; commercial OK <700M MAU)
- Pricing
- Free for self-hosting (Llama community license, <700M MAU); hosted via Together $0.20-$3/M (Maverick); AWS Bedrock + Azure AI consumption
- Funding
- Meta (META) public; ~$1.5T market cap; AI infra capex $60-65B 2025
Taxonomy
- storage
- parametric
- retrieval
- parametric-recall
- persistence
- parametric-permanent
- update
- read-only
- unit
- weight
- governance
- opaque
- conflict
- n/a
When to use
Optimised for: open-weights frontier-tier deployers; cost-efficient inference via MoE; on-prem regulated deployments
Anti-fit: 700M MAU license trigger; not for users wanting weight transparency on dataset (Meta non-disclosing); Behemoth requires extreme GPU (2T MoE)
Pros & cons
Pros
Open weights (community license) — only frontier-tier family runnable on-prem; MoE architecture cheap to serve (17B active params); 10M context (Scout) matches Gemini 3.
Cons
License excludes 700M+ MAU products; dataset opacity; multimodal less mature than GPT-4o / Gemini 3; Behemoth not yet released as of 2025-12.
Claims & capabilities
Llama 4 Scout 109B MoE / 17B active / 10M context; Maverick 400B MoE / 17B active / 1M context; multimodal-native (text+image); Llama community license; first frontier open-weights to use MoE architecture
Technical surface
- API surface
- HTTP via Together / Bedrock / Azure; native via HuggingFace Transformers, vLLM, llama.cpp
- Backend storage
- not applicable — substrate foundation model
- Deployment
- Self-hostable (open-weights) + hosted via AWS Bedrock + Azure AI + Together AI + Groq + Fireworks + Replicate + Databricks
- Embedding model
- not applicable — not a memory product
- Multi-tenancy
- not applicable — substrate foundation model
- MCP
- not applicable — substrate foundation model
- A2A
- not applicable — substrate foundation model
- OpenTelemetry
- not applicable — substrate foundation model
Compare Meta Llama 4 family with…
Similar systems
Other foundation models (substrate reference) in the catalog, ranked by inbound references.
- Anthropic Claude (foundation models) T1
Claude family of frontier foundation models — Sonnet 4 (workhorse), Opus 4.5 (frontier capability), Haiku 4.5 (fast/cheap). Frontier model used as substrate by Claude Code, Anthropic Memory tool, Cline, Aider, Continue.dev, Cursor, Goose, and dozens of other catalog entries. Memory in Claude is parametric (in weights) — discrete episodic memory lives outside the model in the Claude Memory tool (see Platform-provider memory section).
- OpenAI GPT family (GPT-5 / GPT-4o / o3 / o4) T1
OpenAI's foundation model family — GPT-5 (2025-08 flagship), GPT-4o (multimodal workhorse), o3/o4 reasoning series. Substrate for ChatGPT, the OpenAI API, Microsoft Copilot, Azure OpenAI, and many downstream agents in this catalog (Devin, Codex CLI, Operator, Cursor mode). Parametric memory in weights; discrete user memory ships in ChatGPT Memory feature (separate row).
- Google Gemini 3 family T1
Google DeepMind's Gemini 3 frontier model family — Gemini 3 Pro (top tier), 3 Flash (fast), 3 Nano (on-device). Substrate for Google Workspace AI, Vertex AI Agent Platform, Gemini Code Assist, Project Mariner, and Project Astra. Multimodal-native; 1M-10M context window. Largest context window of any frontier model. Used by Gemini Enterprise Agent Platform Memory Bank (separate row).
- Mistral Large 2 / Mixtral family T1
French frontier-model lab (Paris). Family includes Mistral Large 2 (123B dense, Jul-2024), Mixtral 8x22B (open-weights MoE), Mistral Small 3 (24B, open-weights Jan-2025), Codestral (code specialist). EU-headquartered alternative to US/China model labs; GDPR-friendly data residency. Distinct from the Mistral Le Chat memory feature (already in catalog) — this row covers the model family itself as substrate.
- xAI Grok 4 T1
Elon Musk's xAI frontier model — Grok 4 launched 2025-07. Trained on Colossus (100k+ GPU supercluster in Memphis). Deeply integrated with X/Twitter (real-time tweet data access). Grok 4 Heavy variant for premium tier. No frontier-tier open-weights — Grok 1 (314B MoE) is the only weight release.
- 01.AI Yi family T2
Kai-Fu Lee's 01.AI (founded 2023, Beijing). Yi family open-weights — Yi-34B (2023), Yi-Large (proprietary 2024), Yi-Lightning (2024-10, frontier-tier, matched GPT-4o on LMSYS chatbot arena). Apache 2.0 (Yi-34B and base sizes). Important second Chinese open-weights option alongside DeepSeek + Qwen.
Related systems
References (2)
- Azure Machine Learning depends on at runtime — adjacent-infrastructure cell: HuggingFace Transformers; vLLM; llama.cpp; Together AI; Groq; AWS Bedrock; Azure AI Foundry
- Bedrock AgentCore (AWS) depends on at runtime — adjacent-infrastructure cell: HuggingFace Transformers; vLLM; llama.cpp; Together AI; Groq; AWS Bedrock; Azure AI Foundry
Referenced by (2)
- Diffbot depends on at runtime — AG-fine-tuned model based on open-source Llama 3.3.
- Highlight AI depends on at runtime — adjacent-infrastructure cell: Anthropic / OpenAI APIs; local Llama option