Llama 3.1 8B Instruct

Llama 3.1 8B Instruct is the smallest instruction-tuned model in Meta's Llama 3.1 family, released on July 23, 2024. The model card lists 8B parameters, a 128k-token context length, a December 2023 knowledge cutoff and pretraining on more than 15 trillion tokens of data. The model is aimed at assistant-style chat. It remains available on Hugging Face as of October 29, 2026, though newer Llama generations exist; see llama.com for Meta's current models.

Key Features

• Size and context: 8B parameters with a 128k-token context window.
• Languages: English, German, French, Italian, Portuguese, Hindi, Spanish and Thai are the eight supported languages. Other languages require extra fine-tuning and are listed as out of scope.
• License: the Llama 3.1 Community License, a custom commercial license, not an OSI open-source license. The repository is gated and requires accepting Meta's terms. The license permits using outputs for synthetic data generation and distillation.
• Vendor benchmarks: the card reports, for the 8B Instruct model, MMLU 69.4, IFEval 80.4, ARC-C 83.4, GPQA 30.4, HumanEval 72.6, GSM-8K (CoT) 84.5, MATH (CoT) 51.9 and BFCL 76.1. These are Meta's numbers.

Use Cases

• Low-cost chat and assistant features that must run on a single GPU or a laptop with quantization.
• Fine-tuning for domain tasks, or distilling from larger Llama models.
• Multilingual assistants across the eight supported languages.

Limitation: the card's out-of-scope uses include violating the Acceptable Use Policy and using unsupported languages without additional fine-tuning. Larger siblings generally score higher in Meta's own tables.

Pricing and availability

The weights are free to download under the community license. Hosted prices vary by provider, and this entry quotes none because Meta's model card lists no price.

Getting Started

1. Request access on the Hugging Face model page after agreeing to Meta's license.
2. Load meta-llama/Llama-3.1-8B-Instruct with Transformers or a compatible runtime and use the chat template.
3. Compare it with the 70B and 405B variants if quality matters more than cost.

FAQ

What is Llama 3.1 8B Instruct's context length?

128k tokens, per the model card.

Is Llama 3.1 8B Instruct free for commercial use?

It is licensed under the Llama 3.1 Community License, which allows commercial use under its conditions. Read the terms and the Acceptable Use Policy first.

What languages does it support?

English, German, French, Italian, Portuguese, Hindi, Spanish and Thai.

Are there larger Llama 3.1 models?

Yes: Llama 3.1 70B Instruct and Llama 3.1 405B Instruct.

Alternatives

• Llama 3.1 70B Instruct: the mid-size sibling.
• Llama 3.1 405B Instruct: the largest sibling.
• Llama 3.2 1B Instruct: a much smaller Meta model.

Conclusion

Llama 3.1 8B Instruct is a long-context, multilingual small model with a permissive but custom license. It is a solid base for fine-tuning, though newer Meta models exist. See more in the models category.