Model
Model explorer

Llama 3 8B

OPEN
Meta · Llama 3 family · released Apr 18, 2024

First Llama 3 release; 15T-token pretrain, 8K context, best-in-tier 8B at launch (English-only).

ReasoningCodingVisionFunction callingTool useAgentic
486.3
Elo · rank #307
Parameters
8.03B
Active params
8.03B (dense)
Context
8K tokens
Architecture
Dense decoder-only Transformer (GQA)
License
Llama 3 Community License Agreement
Languages
1+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
AGIEvalReasoning45.9%#25
best: OLMo 3-Think 32B · 88.2%
ARC-ChallengeReasoning78.6%#41
best: Llama 3.1 405B · 96.9%
BIG-Bench HardReasoning61.1%#78
best: ERNIE 4.5 300B-A47B · 94.3%
DROPReasoning58.4#52
best: Hunyuan-T1 · 93.1
GPQA DiamondReasoning34.2%#245
best: GPT-5.6 · 94.6%
GSM8KMath79.6%#86
best: Llama 3.1 405B · 96.8%
HumanEvalCoding62.2%#98
best: Claude Opus 4.5 · 99.4%
IFEvalReasoning40.9%#136
best: Gemma 4 26B A4B · 98.5%
MATH-500Math30.0%#165
best: GPT-5 · 99.4%
MiniF2F (valid)Math25.4%#3
best: DeepSeek-Prover-V2 671B · 90.6%
MMLU (EU-21 languages)Knowledge54.8%#8
best: Llama 3.1 70B · 77.1%
MMLU-ProKnowledge35.9%#157
best: Claude Fable 5 · 91.5%
MMLUKnowledge68.4%#157
best: OpenAI o3 · 92.9%
TriviaQAKnowledge78.5%#14
best: Sarvam-1 (2B) · 90.6%
WinoGrandeReasoning76.1%#57
best: PaLM 2 · 90.9%
Run it locally
VRAM @ Q4
5.5 GB
VRAM @ FP16
16 GB
Fits on (Q4)
RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
~100 tok/s on RTX 4090 (Q4, llama.cpp)
Quantizations
GGUF Q4 · AWQ · GPTQ · MLX
Llama 3 family
Elo progression across releases
Llama 3 70BApr 2024886.4Llama 3 8BApr 2024486.3
Resources
HFmeta-llama/Meta-Llama-3-8B-InstructGHmeta-llama/llama3DOCSllama.com/docs/model-cards-and-prompt-formats/meta-llama-3APIAvailable via Meta API
API price weights · each benchmark row carries its own source badge (see methodology)