Model

Llama 3.1 70B

OPEN

Meta · Llama 3.1 family · released Jul 23, 2024

128K-context 70B for large-scale AI-native applications, with built-in multilingual tool use.

ReasoningCodingVisionFunction callingTool useAgentic

980.8

Elo · rank #216

Parameters

70.6B

Active params

70.6B (dense)

Context

128K tokens

Architecture

Dense decoder-only Transformer (GQA)

License

Llama 3.1 Community License Agreement

Languages

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

ARC-ChallengeReasoning94.8%#9

best: Llama 3.1 405B · 96.9%

Arena EloHuman preference1198#41

best: Claude Fable 5 · 1505

BIG-Bench HardReasoning55.9%#84

best: ERNIE 4.5 300B-A47B · 94.3%

BFCLAgents84.8%#7

best: SmolLM3 3B (No Thinking) · 92.3%

Gaokao MathMath41.7%#7

best: Doubao-Seed-1.6 · 96.0%

GPQA DiamondReasoning46.7%#199

best: GPT-5.6 · 94.6%

GSM8KMath95.1%#8

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning86.7%#23

best: Claude 3 Opus · 95.4%

HumanEvalCoding80.5%#55

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning87.5%#40

best: Gemma 4 26B A4B · 98.5%

LogicKorHuman preference7.78#16

best: GPT-4o · 9.33

MATH-500Math68.0%#100

best: GPT-5 · 99.4%

MBPP+Coding86.0%#5

best: Llama 3.1 405B · 88.6%

MGSMMath86.9%#20

best: OpenAI o4-mini · 93.7%

MMLU (EU-21 languages)Knowledge77.1%#1

best: this model · 77.1%

MMLU-ProKnowledge66.4%#102

best: Claude Fable 5 · 91.5%

MMLUKnowledge83.6%#69

best: OpenAI o3 · 92.9%

TruthfulQAKnowledge59.9%#24

best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%

WinoGrandeReasoning85.4%#10

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

40 GB

VRAM @ FP16

140 GB

Fits on (Q4)

M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Q4_K_M weights (~40GB) exceed a single RTX 4090's 24GB VRAM; requires CPU offload, yielding only ~8-12 tok/s degraded throughput rather than a clean in-VRAM figure

Quantizations

GGUF Q4 · AWQ · GPTQ · MLX

Llama 3.1 family

Elo progression across releases

Llama 3.1 405BJul 20241237.4 Llama 3.1 70BJul 2024980.8 Llama 3.1 8BJul 2024553.8

Resources

HFmeta-llama/Meta-Llama-3.1-70B-Instruct GHmeta-llama/llama-models DOCSllama.com/docs/model-cards-and-prompt-formats/llama3_1 APIAvailable via Meta API

API price weights · each benchmark row carries its own source badge (see methodology)