Back to all models
P

Phi-3.5 Mini

MicrosoftReleased 2024-08-21
Official Page

Microsoft's highly efficient small model that punches above its weight class, demonstrating the power of quality synthetic data training.

Overall Score
66%
Cost Efficiency
100
Context Window
128K

Strengths

  • ✓Tiny 3.8B parameters
  • ✓Runs on edge devices
  • ✓Surprisingly capable
  • ✓Fully open-source

Weaknesses

  • ✗Limited reasoning ability
  • ✗Small scale
  • ✗Not competitive with larger models
Architecture
Transformer (Dense)
Parameters
3.8B
Input
$0/M
Output
$0/M

Benchmark Performance

BenchmarkDescriptionScoreCategoryBar
MMLUMassive Multitask Language Understanding80.3%knowledge
HumanEvalCode generation accuracy79%coding
MATHMathematical reasoning75.6%math
GSM8KGrade school math80.1%math
GPQAGraduate-level science Q&A31.3%reasoning
HellaSwagCommon sense reasoning80.1%reasoning
SWE-benchSoftware engineering problems16.9%coding
ARC-CAI2 Reasoning Challenge87%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.