SAHARA Benchmark

[ingest-scout] The most comprehensive African NLP evaluation benchmark to date — 517 African languages and varieties, 16 tasks (classification, NER, QA, reasoning, generation), published at ACL 2025. Created by UBC-NLP to systematically assess LLM performance on African languages and expose policy-induced data inequities. HuggingFace dataset card was fetched and confirmed active. Distinct from all existing registry datasets (MasakhaNER, AfriSenti, etc.) in its continent-wide scale and multi-task scope.

Category

Datasets

Pricing

Not listed

Country

🌍 Pan-African

Last verified

10 Oct 2026

{
  "name": "SAHARA Benchmark",
  "slug": "sahara-benchmark",
  "category": "DATASET",
  "country": "Pan-African",
  "docs_status": "LIVE",
  "licensing_required": "NONE",
  "verified": false,
  "last_verified": "2026-10-10",
  "website": "https://huggingface.co/datasets/UBC-NLP/sahara_benchmark",
  "documentation_url": null
}