Model

Code Llama 34B Python

OPEN

Meta · Code Llama family · released Aug 24, 2023

Python-specialized Code Llama 34B, further trained on 100B Python tokens; HumanEval 53.7%, MBPP 56.2%.

ReasoningCodingVisionFunction callingTool useAgentic

100.1

Elo · rank #385

Parameters

34B

Active params

34B (dense)

Context

16K tokens

Architecture

Dense transformer (decoder-only), Python-specialized, no infilling

License

Llama 2 Community License

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

ARC-ChallengeReasoning50.4%#121

best: Llama 3.1 405B · 96.9%

GSM8KMath32.3%#151

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning76.4%#105

best: Claude 3 Opus · 95.4%

HumanEvalCoding53.7%#114

best: Claude Opus 4.5 · 99.4%

MBPPCoding56.2%#62

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

MMLUKnowledge49.1%#235

best: OpenAI o3 · 92.9%

TruthfulQAKnowledge41.4%#76

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

WinoGrandeReasoning71.9%#88

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

22 GB

VRAM @ FP16

68 GB

Fits on (Q4)

RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

GGUF Q4 · GPTQ · AWQ · MLX

Code Llama family

Elo progression across releases

Code Llama 13BAug 2023232.3 Code Llama 34BAug 2023264.5 Code Llama 34B PythonAug 2023100.1 Code Llama 7BAug 2023127.6 Code Llama 70BJan 2024472.7 Code Llama 70B InstructJan 2024618.8

Resources

HFmeta-llama/CodeLlama-34b-Python-hf GHfacebookresearch/codellama DOCSai.meta.com/blog/code-llama-large-language-model-coding APIAvailable via Meta API

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