Model

Code Llama 70B Instruct

OPEN

Meta · Code Llama family · released Jan 29, 2024

Instruction-tuned 70B coder; 67.8% HumanEval pass@1 edged out GPT-4's zero-shot score at the time, among the top open coding models in early 2024.

ReasoningCodingVisionFunction callingTool useAgentic

618.8

Elo · unrated

Parameters

70B

Active params

70B (dense)

Context

16K tokens

Architecture

Dense transformer decoder (Llama 2 base), instruction fine-tuned

License

Llama 2 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

HumanEval+Coding65.9%#22

best: Mistral Small 3.2 (24B) · 92.9%

HumanEvalCoding67.8%#86

best: Claude Opus 4.5 · 99.4%

MBPPCoding62.2%#54

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

Run it locally

VRAM @ Q4

40 GB

VRAM @ FP16

140 GB

Fits on (Q4)

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

~18 tok/s on RTX 4090 (Q4, llama.cpp)

Quantizations

GGUF Q4 · AWQ · GPTQ · MLX

Code Llama family

Elo progression across releases

Resources

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