Back to all models
L

Llama 3.3 70B

MetaReleased 2024-12-06
Official Page

Meta's efficient 70B instruction model that matches or exceeds Llama 3.1 70B across key benchmarks with improved inference efficiency.

Overall Score
77%
Cost Efficiency
100
Context Window
128K

Strengths

  • ✓Efficient 70B architecture
  • ✓Matches 70B predecessor
  • ✓Fully open-source
  • ✓Better inference efficiency

Weaknesses

  • ✗Smaller than 405B variant
  • ✗Text-only
Architecture
Transformer (MoE)
Parameters
70B
Input
$0/M
Output
$0/M

Benchmark Performance

BenchmarkDescriptionScoreCategoryBar
MMLUMassive Multitask Language Understanding86.1%knowledge
HumanEvalCode generation accuracy88.4%coding
MATHMathematical reasoning87.8%math
GSM8KGrade school math91.5%math
GPQAGraduate-level science Q&A47.8%reasoning
HellaSwagCommon sense reasoning86.5%reasoning
SWE-benchSoftware engineering problems35.2%coding
ARC-CAI2 Reasoning Challenge94%reasoning
AI
ModelBench

The open benchmark platform for comparing the latest AI models and their capabilities.

Platform

  • Compare Models
  • All Models
  • Methodology

AI Labs

  • OpenAI
  • Anthropic
  • Google DeepMind
  • Meta AI

Resources

  • Documentation
  • Benchmark Guide
  • API Reference
Benchmarks are sourced from official model cards and independent evaluations. Data is for informational purposes only.
© 2026 ModelBench. Built for the AI community.