Phi-3.5-mini-instruct

microsoft

Hugging Face ↗

Capability

percentile index

Downloads

726.4K

on HF

Likes

984

Avg score

44.9

9 benchmarks

Weights

Open

Released

Aug 16, 2024

About

Phi-3.5-mini-instruct is a open-weights model in the Phi family from microsoft. Benchmarked on 9 evals, averaging 44.9.

mittext-generation

Benchmark scores · 9

avg 44.9 — higher is better

GSM8Kcurated · 2026-05

MMLUcurated · 2026-05

HumanEvalcurated · 2026-05

IFEvalOpen LLM LB · 2026-05

57.7

BIG-Bench HardOpen LLM LB · 2026-05

36.7

MMLU-ProOpen LLM LB · 2026-05

32.9

GPQA Diamondcurated · 2026-05

MATH Lvl 5Open LLM LB · 2026-05

19.6

MuSROpen LLM LB · 2026-05

10.1

The research behind this model

all papers mentioning it →

Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone Phi-3 Safety Post-Training: Aligning Language Models with a "Break-Fix" Cycle CLIcK: A Benchmark Dataset of Cultural and Linguistic Intelligence in Korean

Details

Publishermicrosoft

FamilyPhi

Parameters—

Licensemit

WeightsOpen

ReleasedAug 16, 2024

History

full ledger →

Capability over time

Append-only ledger — every observed change to this model's numbers.