Model
Model explorer

Llama 3.3 70B

OPEN
Meta · Llama 3.3 family · released Dec 6, 2024

Text-only 70B matching Llama 3.1 405B quality on many benchmarks at roughly 4-5x lower serving cost.

ReasoningCodingVisionFunction callingTool useAgentic
1236.0
Elo · rank #174
Parameters
70B
Active params
70B (dense)
Context
128K tokens
Architecture
Dense transformer decoder with grouped-query attention, 128K context
License
Llama 3.3 Community License Agreement
Languages
8+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
BFCLAgents77.3%#14
best: SmolLM3 3B (No Thinking) · 92.3%
GPQA DiamondReasoning50.5%#188
best: GPT-5.6 · 94.6%
HumanEvalCoding88.4%#28
best: Claude Opus 4.5 · 99.4%
IFEvalReasoning92.1%#10
best: Gemma 4 26B A4B · 98.5%
MATH-500Math77.0%#73
best: GPT-5 · 99.4%
MBPP+Coding87.6%#3
best: Llama 3.1 405B · 88.6%
MGSMMath91.1%#7
best: OpenAI o4-mini · 93.7%
MMLU-ProKnowledge68.9%#95
best: Claude Fable 5 · 91.5%
MMLUKnowledge86.0%#47
best: OpenAI o3 · 92.9%
Run it locally
VRAM @ Q4
42 GB
VRAM @ FP16
140 GB
Fits on (Q4)
M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
70B exceeds a single RTX 4090's 24GB VRAM even at Q4 (~42GB); no verified single-4090 llama.cpp benchmark found.
Quantizations
GGUF · AWQ · GPTQ · MLX
Llama 3.3 family
Elo progression across releases
Llama 3.3 70BDec 20241236.0
Resources
HFmeta-llama/Llama-3.3-70B-InstructGHmeta-llama/llama-modelsDOCSllama.com/docs/model-cards-and-prompt-formats/llama3_3APIAvailable via Meta API
API price weights · each benchmark row carries its own source badge (see methodology)