Atlas

Benchmarks

← All benchmarks

MGSM Swahili

Math · 2022-10-06

Swahili language split of the MGSM multilingual grade-school math benchmark.

Top models (higher is better)

ModelScore
Opus 4.597.2
Gemini 3 Flash Preview97.2
Gemini 3 Pro Preview97.2
GPT-5.296.4
GPT-5.195.6
o395.6
Opus 4.195.2
Gemini 2.5 Flash Preview (09-2025)95.2
Claude 3.7 Sonnet94.8
Sonnet 4.594.8
GPT-594.4
Sonnet 494.0
Claude 3.5 Sonnet (Oct 2024)93.2
Opus 493.2
Grok 4.1 Fast93.2
O193.2
O4 Mini93.2
DeepSeek-V3.292.8
GPT-4.192.8
Grok 3 Mini92.0
Llama 4 Maverick Instruct92.0
Haiku 4.591.6
GPT-4o (2024-08-06)91.6
GPT-4o (2024-11-20)91.6
GPT-5 Mini91.6
Grok 391.6
Kimi K2 Thinking91.6
GLM 4.691.2
DeepSeek-R190.8
Gemini 2.5 Flash-Lite Preview (09-2025)90.8
Grok 4 Fast90.8
Gemini 2.0 Flash 00190.4
Qwen3-235B-A22B90.4
GLM-4.789.6
Llama 4 Scout Instruct89.6
Qwen3 Max (rolling alias)89.6
GLM-4.589.2
Llama 3.3 70B Instruct89.2
o3-mini89.2
gpt-oss-120b88.8
Loading Atlas data…