live
weekly refresh
basedagi.org
benchmark evidence

AI Language Proficiency Monitor Arabic

Mean ARC, MGSM, and MMLU accuracy for Arabic, from the AI Language Proficiency Monitor.

winner on AI Language Proficiency Monitor Arabic
direct benchmark result, not a broad vertical composite | source row dated 2026-05-19
scored on 2026-05-19
latest mapped results | top 20
#ModelScoreEvidenceTested
1Qwen: Qwen3 30B A3B
Qwen
90.5
model-only
independent_benchmark
2026-05-19
2Anthropic: Claude Opus 4.6
Anthropic
90.0
model-only
independent_benchmark
2026-05-19
3Anthropic: Claude Opus 4.5
Anthropic
90.0
model-only
independent_benchmark
2026-05-19
4Anthropic: Claude Sonnet 4.6
Anthropic
88.9
model-only
independent_benchmark
2026-05-19
5OpenAI: GPT-5
Openai
88.9
model-only
independent_benchmark
2026-05-19
6Anthropic: Claude Haiku 4.5
Anthropic
86.7
model-only
independent_benchmark
2026-05-19
7OpenAI: GPT-5.2
Openai
86.7
model-only
independent_benchmark
2026-05-19
8OpenAI: GPT-5.1
Openai
86.7
model-only
independent_benchmark
2026-05-19
9Anthropic: Claude Sonnet 4.5
Anthropic
86.7
model-only
independent_benchmark
2026-05-19
10Google: Gemini 2.5 Pro
Google
86.7
model-only
independent_benchmark
2026-05-19
11Anthropic: Claude Sonnet 4
Anthropic
83.3
model-only
independent_benchmark
2026-05-19
12OpenAI: GPT-5.4
Openai
83.0
model-only
independent_benchmark
2026-05-19
13Amazon: Nova Pro 1.0
Amazon
80.0
model-only
independent_benchmark
2026-05-19
14OpenAI: GPT-4.1
Openai
76.7
model-only
independent_benchmark
2026-05-19
15Meta: Llama 3.3 70B Instruct
Meta Llama
76.7
model-only
independent_benchmark
2026-05-19
16Meta: Llama 4 Maverick
Meta Llama
73.3
model-only
independent_benchmark
2026-05-19
17Google: Gemini 2.5 Flash
Google
43.3
model-only
independent_benchmark
2026-05-19
18Mistral: Mistral Nemo
Mistralai
26.7
model-only
independent_benchmark
2026-05-19
what this result means

Mean ARC, MGSM, and MMLU accuracy for Arabic, from the AI Language Proficiency Monitor.

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

A win here is a win on AI Language Proficiency Monitor Arabic. Broad task pages require independent corroboration before naming a general winner.

source record
category: multilingual
metric: accuracy
matched models: 18
latest source date: 2026-05-19
direction: higher is better
inspect upstream source ->