Llama (Meta)

Category: AI Platform

Open-weight models for building on, not a marketing tool.

Llama is Meta's family of open-weight language models. The weights are free to download under the Llama Community Licence, which is not an OSI-approved open-source licence, and the real cost is whatever you pay to host them or to an inference provider. The latest verified family is Llama 4, with several model sizes. For marketers this is infrastructure rather than a tool: you would encounter it inside something else, or use it if you were building your own application and wanted to run a model on your own hardware for cost or data-control reasons. Meta has since pivoted its flagship work to the proprietary Muse models (llama.com now redirects to Meta's Muse developer site), so Llama is no longer where the company's new development is heading and existing Llama weights are effectively a legacy option. There is no Llama MCP server; the associated stack consumes MCP tools rather than publishing one.

Our verdict

Worth knowing about rather than using directly, and we would be clear that this belongs in a developer category rather than next to marketing software. The genuine use case for a marketing team is data control: if you cannot send client material to a third-party API, running an open-weight model on infrastructure you control solves that in a way no commercial assistant does, and Llama 4 is capable enough for most text work. Two licence details worth knowing. It is not open source in the OSI sense despite being widely described that way, and the licence includes a clause triggered above a very large monthly user count. More relevantly for European readers, the Llama 4 licence withholds multimodal rights from individuals domiciled in the EU and companies headquartered there. UK entities sit outside that wording and so are not caught, but check the specific version's licence rather than assuming. Claims of a Llama 5 are unverified, and with Meta now putting its newest models behind the proprietary Muse line, treat Llama as a stable but legacy choice rather than something that will keep improving.

Best for: Technical teams building their own applications who want to self-host a model for cost or data-control reasons.

Pros

  • Weights are free to download and can be self-hosted for full data control
  • Capable enough for most text work without a third-party API
  • The practical answer if you cannot send client material to a commercial API
  • Large ecosystem of tooling and fine-tunes
  • UK entities are outside the EU multimodal restriction in the licence

Cons

  • Not a marketing tool, this is model infrastructure for developers
  • Not open source in the OSI sense, despite being widely described that way
  • The licence withholds multimodal rights from EU-domiciled individuals and companies
  • Real cost is hosting or inference provider fees, which are not trivial
  • No MCP server, and the associated stack is a client rather than a server
  • Meta has pivoted to proprietary Muse models, so Llama is a legacy option with an uncertain future

Pricing: Weights free under licence, cost is hosting (Free)

Rating: 4/5