Meta Llama 3.1 now available on Workers AI

Meta’s Llama 3.1 arrives on Workers AI

Cloudflare has added Meta’s Llama 3.1 8B model to its serverless inference platform, making it available to all Workers AI users immediately. The model is accessible by changing the model ID to @cf/meta/llama-3.1-8b-instruct or via the Workers AI Playground, and it remains free during its beta period.

The Llama 3.1 release brings several technical upgrades over previous iterations. The model operates at bfloat16 precision, includes native function calling, and expands language coverage to eight languages: English, French, German, Hindi, Italian, Portuguese, Spanish, and Thai. Developers can prompt the model and receive responses in those languages without additional configuration, allowing a single model to serve a broader global user base.

Function calling built in

Unlike earlier Llama versions, 3.1 supports function calling out of the box. The model can generate structured JSON outputs that feed directly into APIs, removing the need for a fine-tuned variant specialized for tool use. This consolidation means developers can handle multiple tasks, from general conversation to API orchestration, with one model.

Workers AI’s embedded function calling feature works with Llama 3.1 as well. That feature reduces the number of manual inference requests by leveraging Cloudflare Workers and the open-source ai-utils package to orchestrate the back-and-forth between the model and external tools. The utility package also provides helper methods that auto-generate tool schemas. A practical example writes key-value pairs to Workers KV following a model-generated function call.

Availability and usage requirements

Cloudflare positions this release as part of its broader commitment to open-source models, with a focus on developer tooling and distributed GPU infrastructure. Llama models have consistently delivered strong results in general knowledge, math, steerability, tool use, and multilingual translation, and Workers AI will continue distributing the collection.

Developers should review Meta’s Acceptable Use Policy and License before deploying the model in production. Full setup guidance is available in the Workers AI developer documentation.