// BENCHMARK

MGSM benchmark

AI model leaderboard for the MGSM benchmark. Compare how large language models score on MGSM, see the full ranking, and understand what this AI benchmark measures. GPT-4o mini currently leads with 81.7. Multilingual GSM8K — translations of grade-school math problems into 10 languages.

Leaderboard

# Model Organization Score Variant Source
#1 GPT-4o mini OpenAI 81.7 cited-phi35moe official ↗
#2 Gemini 1.5 Flash (Sep 2024) Google DeepMind 75.8 cited-phi35moe official ↗
#3 Gemma 2 9B Google DeepMind 75.1 instruct-cited-phi35moe official ↗
#4 Qwen2.5-7B Qwen 64.5 cited-phi4-mini official ↗
#5 Phi-4 Mini Microsoft 63.9 instruct-0-shot-cot official ↗
#6 Mistral NeMo Mistral AI 63.3 instruct-cited-phi35moe official ↗
#7 Phi-3.5-MoE Microsoft 58.7 instruct-0-shot-cot official ↗
#8 Llama 3.1-8B Meta AI 56.7 instruct-cited-phi35moe official ↗
#9 Qwen2.5-3B Qwen 53.5 cited-phi4-mini official ↗
#10 phi-3.5-mini Microsoft 49.6 cited-phi4-mini official ↗
#11 Granite-4.0-H-Tiny IBM 45.36 8-shot official ↗
#12 Ministral 8B Mistral AI 44.6 cited-phi4-mini official ↗
#13 Llama 3.2 3B Meta AI 44.6 cited-phi4-mini official ↗
#14 Granite-4.0-H-Micro IBM 44.48 8-shot official ↗
#15 Granite-4.0-H-Small IBM 38.72 8-shot official ↗

Frequently asked questions about MGSM

What is the MGSM benchmark?

Multilingual GSM8K — translations of grade-school math problems into 10 languages.

How is the MGSM benchmark scored?

MGSM is scored using the accuracy metric, where a higher score is better. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.

Which AI model scores highest on MGSM?

As of the latest reported scores on GenAIList, GPT-4o mini achieves the highest result on MGSM with a score of 81.7.

Is a higher MGSM score better?

Yes. On MGSM a higher score indicates better performance, so models near the top of the leaderboard are the strongest.