Llama 3.1

Meta's 405B Llama 3.1 base model — the largest openly released foundation model for fine-tuning and research at frontier scale.

Meta's 405B Llama 3.1 flagship — the largest openly released model at launch, competitive with GPT-4o on most benchmarks.

Meta's 70B Llama 3.1 instruct model — the sweet spot of the 3.1 family, with 128K context and near-GPT-4-level performance.

Meta's 8B Llama 3.1 base model — strong foundation for fine-tuning with a 128K context window.

Meta's Llama 3.1 8B instruction-tuned model with 128K context, strong general assistant.