LLaMA 3 (8B/70B)Open Weight

Released

LLaMA 3 (8B/70B) is an AI model released by Meta on Thursday, Apr 18 2024, 80 days after Code Llama 70B. It is an open-weight model — the trained weights are available to download and run. It was released in 8B and 70B parameter sizesThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality., with a 8k token context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider.. Benchmark results (shown below) cover HumanEval, MATH, MMLU, and GSM8K.

Benchmarks

Coding

HumanEvalFunction synthesis — 164 small Python problems: the AI is given a function's description and has to write the working function. This was the coding benchmark of the GPT-3.5 and GPT-4 era, before the field moved to fixing real bugs in real repositories. Higher is better.
81.7%
#34 of 74

Reasoning & science

MATHReasoning — accuracy on the original competition mathematics test set
50.4%
#32 of 50
MMLUGeneral knowledge — A 57-subject multiple-choice exam — history, law, medicine, maths — that was the standard measure of how much a model knows from 2020 until roughly 2024, when frontier scores crowded into the high 80s and labs moved on to harder tests. The scores here were published years apart under different testing setups, so read them as a historical record rather than a like-for-like ranking. Higher is better.
82%
#29 of 68
GSM8KGrade-school math — Grade-school maths word problems that take a few steps of arithmetic to work through. It separated the models of 2022 and 2023 sharply, then saturated. One caveat on the historical numbers: OpenAI included part of the GSM8K training set in GPT-4's pre-training mix, so GPT-4's score is not a clean few-shot result. Higher is better.
93%
#6 of 39

Compare LLaMA 3 (8B/70B) with

LLaMA 3 (8B/70B)

About

Llama 3 launched on April 18, 2024 with 8B and 70B parameter models, each offered as pretrained and instruction-tuned checkpoints. Meta emphasized improvements in following instructions, coding, reasoning and reducing unnecessary refusals. The first release was text based, and its 8,192-token context window was separate from the much longer context introduced with Llama 3.1 later that year. The publisher benchmarks listed here use the 70B Instruct checkpoint.

Meta reported pretraining on more than 15 trillion tokens and introduced a larger tokenizer vocabulary than Llama 2. Developers could download the models for fine-tuning and self-hosting under the Llama 3 Community License. The launch also powered the updated Meta AI assistant, connecting the research model family to a consumer product without making the assistant and the raw checkpoints interchangeable.

Frequently asked questions

LLaMA 3 (8B/70B) was released by Meta on Thursday, Apr 18 2024.

All Meta releases

19 tracked