live
weekly refresh
basedagi.org
benchmark evidence

MGSM (Aggregate)

Multilingual Grade School Math: aggregate accuracy across 10 languages (Spanish, French, German, Russian, Chinese, Japanese, Thai, Swahili, Bengali, Telugu). Score = % of math word problems solved correctly. These entries are OpenAI-published model results, not an independent cross-provider leaderboard.

winner on MGSM (Aggregate)
direct benchmark result, not a broad vertical composite | source row dated 2026-06-07
scored on 2026-06-07
latest mapped results | top 20
#ModelScoreEvidenceTested
1OpenAI: o4 Mini
Openai
93.7
model-only
official_provider
2026-06-07
2Meta: Llama 4 Maverick
Meta Llama
92.3
model-only
independent_benchmark
2000-01-01
3OpenAI: o3
Openai
92.3
model-only
official_provider
2026-06-07
4Meta: Llama 3.3 70B Instruct
Meta Llama
91.1
model-only
independent_benchmark
2000-01-01
5OpenAI: o3 Mini
Openai
90.8
model-only
official_provider
2026-06-07
6Meta: Llama 4 Scout
Meta Llama
90.6
model-only
independent_benchmark
2000-01-01
7OpenAI: o1
Openai
89.3
model-only
official_provider
2026-06-07
8OpenAI: GPT-4.1
Openai
86.9
model-only
official_provider
2026-06-07
9Qwen: Qwen3 235B A22B
Qwen
83.5
model-only
independent_benchmark
2000-01-01
what this result means

Multilingual Grade School Math: aggregate accuracy across 10 languages (Spanish, French, German, Russian, Chinese, Japanese, Thai, Swahili, Bengali, Telugu). Score = % of math word problems solved correctly. These entries are OpenAI-published model results, not an independent cross-provider leaderboard.

This benchmark contributes direct public evidence. Read its scope before generalizing the result.

A win here is a win on MGSM (Aggregate). Broad task pages require independent corroboration before naming a general winner.

source record
category: multilingual
metric: accuracy
matched models: 9
latest source date: 2026-06-07
direction: higher is better
inspect upstream source ->