Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

MGSM

Multilingual Grade School Math (MGSM) benchmark evaluates language models' mathematical reasoning capabilities across multiple languages. Contains 250 carefully translated math word problems derived from the GSM8K dataset, covering 10 typologically diverse languages including underrepresented ones like Bengali, Telugu, and Swahili.
Source:

Model Performance

#1
Claude-4.1-Opus-Thinking
94.8%
#2
Claude-4.1-Opus
94.2%
#3
Claude-4.0-Opus
93.8%
#4
OpenAI-O4-mini-high
93.4%
#5
OpenAI-O4-mini-medium
93.4%
#6
Claude-3.7-Sonnet-Thinking
93.0%
#7
Claude-4.0-Sonnet
93.0%
#8
GPT-5-mini
92.6%
#9
Claude-3.5-Sonnet-1022
92.6%
#10
Qwen-3-Thinking
92.5%
#11
Qwen-3
92.5%
#12
Llama-4-Maverick-17B
92.4%
#13
Qwen-3-Max-Preview
92.4%
#14
Claude-3.7-Sonnet
92.4%
#15
DeepSeek-R1
92.3%
#16
Gemini-2.5-Pro-0325
92.2%
#17
Gemini-2.5-Pro-Thinking
92.2%
#18
DeepSeek-V3
92.1%
#19
OpenAI-O3-high
91.7%
#20
DeepSeek-V3-0324
91.7%
#21
OpenAI-O3-medium
91.7%
#22
Grok-3-Beta
91.3%
#23
OpenAI-O3-mini-medium
91.3%
#24
OpenAI-O3-mini-high
91.3%
#25
Grok-3-Think
91.3%
#26
Llama-3.3-70B
91.1%
#27
Mistral-Medium-3.1
90.9%
#28
Grok-4-Thinking
90.9%
#29
Claude-4.0-Sonnet-Thinking
90.9%
#30
Kimi-K2-Instruct
90.9%
#31
Grok-4
90.9%
#32
Grok-3-Mini-Beta
90.4%
#33
Gemini-2.5-Flash-Thinking
90.0%
#34
Gemini-2.5-Flash
89.8%
#35
OpenAI-O1-1217
89.3%
#36
Gemini-1.5-Pro
89.2%
#37
Gemini-2.0-Flash
89.0%
#38
GPT-4.1
87.7%
#39
Mistral-Large-2
87.2%
#40
GPT-4-mini
86.2%
#41
Cohere-Command-A
85.7%
#42
Claude-3.5-Haiku
84.6%