Qwen2.5-72B-Instruct

Qwen

Hugging Face ↗

Parameters

72B

Downloads

900K

on HF

Likes

942

Avg score

67.2

10 benchmarks

Weights

Open

Released

Sep 16, 2024

About

Qwen2.5-72B-Instruct is a 72B-parameter open-weights model in the Qwen2 family from Qwen. Benchmarked on 10 evals, averaging 67.2.

othertext-generation

Benchmark scores · 10

avg 67.2 — higher is better

GSM8Kcurated · 2026-05

95.8

HumanEvalcurated · 2026-05

86.6

IFEvalOpen LLM LB · 2026-05

86.4

MMLUcurated · 2026-05

86.1

MATHcurated · 2026-05

83.1

BIG-Bench HardOpen LLM LB · 2026-05

61.9

MATH Lvl 5Open LLM LB · 2026-05

59.8

MMLU-ProOpen LLM LB · 2026-05

51.4

GPQA Diamondcurated · 2026-05

MuSROpen LLM LB · 2026-05

11.7

The research behind this model

all papers mentioning it →

YaRN: Efficient Context Window Extension of Large Language Models Qwen2 Technical Report Qwen2.5 Technical Report71 cites · Dec 19, 2024 FormalRewardBench: A Benchmark for Formal Theorem Proving Reward ModelsMay 11, 2026

Mentions matched by name in title/abstract — from the arXiv + HF daily corpus.

Details

PublisherQwen

FamilyQwen2

Parameters72B

Licenseother

WeightsOpen

ReleasedSep 16, 2024

History

full ledger →

Capability over time

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