Llama-3.1-8B-Instruct

weights

open(downloadable)

data

closed

code

partial(no training pipeline)

license

Llama-3.1-Community(non-OSI

Meta releases the weights under the Llama 3.1 Community License, non-OSI and requiring a separate license from Meta above 700 million monthly active users. That cap permits commercial use below the threshold rather than forbidding it, so the release counts as open weights rather than fully open source: the training data and pipeline are unpublished, and the manual access gate on the repository does not itself reduce its openness.

• https://huggingface.co/api/models/meta-llama/Llama-3.1-8B-Instruct recorded 2026-09-29

  API record lists "id":"meta-llama/Llama-3.1-8B-Instruct" with "gated":"manual" and "downloads":6135791.

• https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct recorded 2026-09-29

  Model card and repo tags show `license:llama3.1`; the license body reads "is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta"; the repository is gated ("gated":"manual" in the API record, not on this page), and the card lists no training pipeline.

• https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct recorded 2026-09-29

  Model card: "Llama 3.1 was pretrained on ~15 trillion tokens of data from publicly available sources" and the fine-tuning data "includes publicly available instruction datasets, as well as over 25M synthetically generated examples"; the corpora themselves are not released. Searched the body for a training pipeline, training code or post-training data release: none.

Adoption is read from Hugging Face downloads on the sole declared checkpoint, meta-llama/Llama-3.1-8B-Instruct. No other artifact is declared for this record, so the reading does not draw on any additional distribution surface that might place it higher.

• https://huggingface.co/api/models/meta-llama/Llama-3.1-8B-Instruct recorded 2026-08-13

  7,345,657 downloads in the trailing 30 days for meta-llama/Llama-3.1-8B-Instruct

A competent 2024-generation 8B chat model, a long way below the 2026 frontier and below the larger Llama 3.1 SKUs. The card's benchmark table gives Llama 3.1 8B Instruct MMLU 69.4 (5-shot, macro_avg/acc) against 83.6 for the 70B and 87.3 for the 405B, and publishes no reasoning or agentic frontier scores.

• https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct recorded 2026-09-29

  Benchmark table lists MMLU (5-shot, macro_avg/acc) for Llama 3.1 8B Instruct, 70B and 405B, with a context length of 128k; "The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF)". Searched the body for reasoning or agentic benchmark scores: none.