Atlas

Benchmarks

← All benchmarks

MGSM French

Math · 2022-10-06

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

Top models (higher is better)

ModelScore
Opus 4.193.2
Opus 492.8
Claude 3.5 Sonnet (Oct 2024)92.0
DeepSeek-V391.6
Llama 4 Maverick Instruct91.6
Sonnet 491.2
DeepSeek-R190.8
Grok 390.8
Llama 3.3 70B Instruct90.8
DeepSeek-V3-032490.4
DeepSeek-V3.290.4
Grok 490.4
Opus 4.590.0
O4 Mini90.0
Sonnet 4.589.6
Kimi K2 Instruct88.8
Claude 3.7 Sonnet88.4
Mistral Large 3 675B Instruct 251288.0
Qwen3 Max Preview88.0
Qwen3-235B-A22B87.6
Gemini 3 Flash Preview87.2
Llama 4 Scout Instruct87.2
GPT-4o Mini86.8
GPT-4.1 Mini86.4
Grok 3 Mini86.0
Qwen3 Max (rolling alias)86.0
Command A85.6
Gemini 2.5 Flash-Lite Preview (09-2025)85.6
GPT-5 Nano85.2
GLM-4.584.8
Grok 284.8
gpt-oss-20b84.4
o384.4
Haiku 3.584.0
Haiku 4.584.0
GPT-4o (2024-08-06)84.0
Mistral Large 2.1 (Instruct 2411)84.0
O184.0
Grok 4 Fast83.6
MiniMax M2.183.6
Loading Atlas data…