What is the DataLens Africa LLM Leaderboard?

The DataLens Africa LLM Leaderboard is the definitive benchmark ranking frontier large language models on their ability to understand African languages, cultures, and healthcare contexts. It evaluates models across four African-language benchmarks — AfriMCQA, AfriMMLU, MasakhaNEWS, and AfriMedQA — and is updated continuously as new models are evaluated.

Which LLM performs best on African language benchmarks?

As of September 2026, Gemini 3.7 Flash by Google leads the DataLens Africa LLM Leaderboard with an overall score of 83.39%. It holds the highest single-benchmark score ever recorded on the leaderboard — 91.11% on AfriMCQA — while Gemini 3.5 Flash leads AfriMedQA at 84.71%. Gemini 3.5 Flash ranks second overall at 82.12%, and GPT-5.6 Sol ranks third at 80.33%.

What benchmarks does the DataLens Africa LLM Leaderboard use?

The DataLens Africa LLM Leaderboard evaluates models across four benchmarks: AfriMCQA (multiple-choice question answering across African cultural and factual topics), AfriMMLU (massive multitask language understanding adapted for African contexts), MasakhaNEWS (news topic classification across 16 African languages), and AfriMedQA (medical question answering in African language and healthcare settings).

How is the overall score calculated on the DataLens Africa LLM Leaderboard?

The overall score is a weighted average of a model's scores across the available benchmarks — AfriMCQA, AfriMMLU, MasakhaNEWS, and AfriMedQA. Models that have not been evaluated on a specific benchmark, or whose results did not meet the evaluation validity threshold, are scored on the benchmarks available. All scores are expressed as percentages.

Which LLM is best for African healthcare AI applications?

Based on AfriMedQA scores on the DataLens Africa LLM Leaderboard, Gemini 3.5 Flash leads at 84.71%, followed by Gemini 3.7 Flash at 82.65% and Claude Opus 5 at 79.39%. Both Google's Gemini family and Anthropic's Claude family show the strongest performance on African medical question-answering tasks as of September 2026.

How often is the DataLens Africa LLM Leaderboard updated?

The DataLens Africa LLM Leaderboard is updated continuously as new model evaluations are completed. The first cohort of 10 models was published in February 2026. A second cohort of 5 new models was added in May 2026, bringing the total to 15 ranked models. A third cohort of 6 new models was added in August 2026, bringing the total to 21 ranked models. Grok 4.6 was added in September 2026, bringing the total to 22 ranked models. New evaluations are added on a rolling basis as frontier models are submitted or selected for evaluation.

How do I get my model added to the DataLens Africa LLM Leaderboard?

Submit it through the DataLens African AI Evaluation route at https://datalens.africa/evaluate. Every submission is reviewed by hand and returned with a costed evaluation quote before any work begins. Models evaluated this way are considered for inclusion in the next cohort, and you choose at submission whether your results are published publicly or kept private.

Which African languages are covered by the LLM benchmarks?

The benchmarks used on the DataLens Africa LLM Leaderboard cover a range of African languages. MasakhaNEWS spans 16 African languages including Hausa, Yoruba, Swahili, Amharic, Igbo, and Wolof. AfriMCQA and AfriMMLU cover African factual and cultural knowledge across multiple linguistic contexts. AfriMedQA focuses on medical terminology and clinical scenarios relevant to African healthcare settings.

Measuring AI Intelligence in African Contexts.

The Datalens Africa LLM Leaderboard is the definitive benchmark for evaluating AI understanding of African contexts. Compare how leading AI systems perform across language, culture, knowledge, and healthcare using the most comprehensive African evaluation framework.

Explore the Rankings View the Benchmarks

15k+

Questions

16+

Languages

32

Specialties

5

LLM Families

Rankings

African AI Benchmark Leaderboard

← Swipe to see all benchmark scores →

Rank

Model

AfriMCQA

AfriMMLU

MasakhaNEWS

AfriMedQA

Overall Score

👑1

Gemini 3.7 Flash

Google · Aug '26

91.11%

82.75%

77.06%

82.65%

83.39%

Gemini 3.5 Flash

Google · May '26

90.14%

80.88%

72.74%

84.71%

82.12%

GPT-5.6 Sol

OpenAI · Aug '26

84.38%

78.13%

82.88%

75.93%

80.33%

Grok 4.6

xAI · Sep '26

83.65%

77.50%

74.61%

73.31%

77.27%

Claude Opus 4.6

Anthropic · May '26

82.45%

68.13%

79.19%

78.99%

77.19%

DeepSeek-V4-Pro

DeepSeek AI · May '26

—

76.75%

75.80%

76.24%

76.26%

GPT-5.4

OpenAI · May '26

78.37%

72.75%

77.71%

74.71%

75.88%

DeepSeek-V4-Flash

DeepSeek AI · Aug '26

—

77.75%

79.87%

69.26%

75.63%

GPT-5.1

OpenAI · Feb '26

81.22%

62.13%

78.61%

76.72%

74.67%

Gemini 3.1 Flash Lite

Google · May '26

85.58%

69.38%

71.44%

72.06%

74.61%

Claude Sonnet 5

Anthropic · Aug '26

75.72%

55.50%

82.62%

78.76%

73.15%

Gemini 2.5 Pro

Google · Feb '26

85.58%

79.47%

51.22%

75.95%

73.05%

GPT-5.6 Luna

OpenAI · Aug '26

68.51%

65.37%

81.25%

76.40%

72.88%

DeepSeek-R1

DeepSeek AI · Feb '26

—

70.63%

73.87%

72.99%

72.50%

Claude Sonnet 4.6

Anthropic · Feb '26

79.33%

62.38%

68.75%

78.04%

72.12%

GPT-5.2

OpenAI · Feb '26

75.50%

67.38%

75.27%

69.97%

72.03%

Gemini 2.5 Flash

Google · Feb '26

81.05%

60.50%

71.20%

74.42%

71.79%

Grok 4.1 Fast Reasoning

xAI · Feb '26

72.60%

59.63%

64.90%

70.13%

66.81%

Claude Opus 5

Anthropic · Aug '26

59.37%

51.00%

74.45%

79.39%

66.05%

DeepSeek-V3.2

DeepSeek AI · Feb '26

—

61.50%

64.66%

71.06%

65.74%

Grok 4 Fast Reasoning

xAI · Feb '26

74.76%

54.13%

60.26%

71.32%

65.12%

Claude Haiku 4.5

Anthropic · Feb '26

63.46%

54.75%

67.72%

62.49%

62.10%

Your model here

Submit a model and we return a scored, reproducible evaluation across seven dimensions.

* Scores represent average performance across all benchmark sub-tasks in each dataset. Version 1.4 · Last updated: 1 September 2026.

AfriMCQA

Multilingual Cultural Understanding · Accuracy (%)

AfriMMLU

Knowledge & Reasoning Across Languages · Accuracy (%)

MasakhaNEWS

News Topic Classification · Macro-F1 (%)

AfriMedQA v2

Pan-African Clinical QA · Macro-Accuracy (%)

Benchmarks

Four Pillars of African AI Evaluation

Each benchmark targets a distinct dimension of African contextual understanding, spanning culture, language, and clinical medicine, and providing a multidimensional view of LLM capability.

Afri-MCQA

Multimodal, culturally-grounded multiple-choice questions in 16+ African languages, testing deep cultural and linguistic comprehension.

Culture Multilingual Multimodal

Languages 16+

Task Type MCQ / VQA

Metric Accuracy

AfriMMLU

Human-translated MMLU spanning 5 subjects across 17 African languages. Tests knowledge across geography, law, economics, global facts and mathematics.

Knowledge Reasoning Human-Translated

Subjects 5 Domains

Languages 17

Metric Accuracy

MasakhaNEWS

News topic classification across 16 African languages and 7 categories. Tests in-language understanding of real-world African news across diverse topics.

NLP Classification News

Categories 7 Topics

Languages 16

Metric Macro-F1

AfriMedQA v2

Pan-African medical QA dataset with 15,275 questions from 621 contributors across 16 countries, covering 32+ clinical specialties.

Medical Expert-Curated Healthcare

Questions 15,275

Specialties 32+

Metric Macro-Accuracy

Evaluated Models

Frontier AI Systems Under Review

Google

Gemini Series

Five Gemini models evaluated; Gemini 3.7 Flash leads the entire leaderboard with the highest overall score.

Models3.7 Flash · 3.5 Flash

Best Score83.39%

OpenAI

GPT Series

Five GPT models evaluated across all four African benchmarks; GPT-5.6 Sol is the strongest, leading on MasakhaNEWS.

Models5.1 · 5.2 · 5.4 · 5.6 Sol · 5.6 Luna

Best Score80.33%

xAI

Grok Series

Three Grok models evaluated with web access disabled for fairness; Grok 4.6 jumps the family to #4 overall, well ahead of the Fast Reasoning models.

ModelsGrok 4 · Grok 4.1 · Grok 4.6

Best Score77.27%

Anthropic

Claude Series

Five Claude models evaluated; Opus 4.6 leads the family and ranks #5 overall, with Sonnet 5 close behind on MasakhaNEWS.

ModelsOpus 4.6/5 · Sonnet 4.6/5 · Haiku 4.5

Best Score77.19%

DeepSeek AI

DeepSeek Series

DeepSeek-V4-Pro, V4-Flash, R1, and V3.2 evaluated; V4-Pro leads the family with strong MMLU and MasakhaNEWS scores.

ModelsV4-Pro · V4-Flash · R1 · V3.2

Best Score76.26%

DataLens African AI Evaluation

This leaderboard ranks the models we chose. Get yours measured.

The four benchmarks here are one dimension of seven. Submit your model and we return a scored, reproducible report — per language, with a documented methodology.

Submitting costs nothing and commits you to nothing — we reply with a scoped, costed quote.

African-language task performance — this leaderboard

Token efficiency / cost-to-serve

Cultural appropriateness

Factuality & hallucination rate

Safety in African contexts

Instruction following & reasoning

Dialect & code-switch robustness

DataLens Studio

Help the next model
top this leaderboard.

Every score on this leaderboard is a reflection of the data behind the model. Better African language annotations produce better training data, which trains models that score higher on these benchmarks. DataLens Studio is the platform where that work happens — built specifically for African language labeling, RLHF, and cultural context annotation.

Annotate African language data

Label text, audio, and cultural context across 50+ African languages.

Build better training datasets

Human-verified, culturally-grounded data that frontier models are missing.

Models score higher on African benchmarks

Your contributions directly move the needle on this leaderboard.

DataLens Studio

African Language Annotation Platform

Text, audio & RLHF annotation tasks

50+ African languages supported

Earn while contributing to African AI

Quality-reviewed by African language experts

Start Annotating Sign in to your account

Every annotation is a step toward AI that truly understands Africa.

Measuring AI Intelligence in African Contexts.

15k+

16+

32

5

African AI Benchmark Leaderboard

Four Pillars of African AI Evaluation

Frontier AI Systems Under Review

This leaderboard ranks the models we chose. Get yours measured.

Help the next modeltop this leaderboard.

Help the next model
top this leaderboard.