Model

LLaMA 65B

OPEN

Meta · LLaMA family · released Feb 24, 2023

Flagship of Meta's first LLaMA release; dense 65B model that kicked off the modern open-weights LLM wave despite its gated non-commercial license.

ReasoningCodingVisionFunction callingTool useAgentic

240.9

Elo · rank #358

Parameters

65.2B

Active params

65.2B (dense)

Context

2K tokens

Architecture

Dense decoder-only Transformer

License

LLaMA Non-Commercial Research License (gated application access)

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AGIEvalReasoning47.6%#22

best: OLMo 3-Think 32B · 88.2%

ARC-ChallengeReasoning56.0%#103

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning78.9%#28

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning43.5%#108

best: ERNIE 4.5 300B-A47B · 94.3%

GSM8KMath50.9%#132

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning84.2%#53

best: Claude 3 Opus · 95.4%

HumanEvalCoding23.7%#163

best: Claude Opus 4.5 · 99.4%

MATH-500Math10.6%#192

best: GPT-5 · 99.4%

MBPPCoding37.7%#88

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLUKnowledge63.4%#188

best: OpenAI o3 · 92.9%

OpenBookQAReasoning60.2%#14

best: Claude 1 · 90.8%

PIQAReasoning82.8%#21

best: GPT-4o mini · 93.1%

RACE-HReasoning51.6%#11

best: Claude 3 Opus · 92.9%

Social IQaReasoning52.3%#20

best: Apple DCLM-Baseline 7B · 82.9%

TriviaQAKnowledge68.2%#26

best: Sarvam-1 (2B) · 90.6%

TruthfulQAKnowledge48.7%#59

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

WinoGrandeReasoning77.0%#51

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

37 GB

VRAM @ FP16

130 GB

Fits on (Q4)

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

Q4 weights (~37GB) exceed the RTX 4090's 24GB VRAM, requiring CPU/disk offload; no reliably sourced single-GPU 4090 tokens/sec figure exists for this checkpoint.

Quantizations

GGUF · GPTQ · AWQ · GGML

LLaMA family

Elo progression across releases

LLaMA 13BFeb 20230.6 LLaMA 33BFeb 2023184.9 LLaMA 65BFeb 2023240.9 LLaMA 7BFeb 2023-180.9

Resources

HFhuggyllama/llama-65b GHmeta-llama/llama DOCSai.meta.com/blog/large-language-model-llama-meta-ai APIAvailable via Meta API

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