nsanku-MMLU - Kasem: leaderboard

Accuracy on the Kasem questions of GhanaNLP's nsanku-MMLU, four-option multiple-choice items from Ghana's National Teaching Council licensing practice tests for that language (100 or 200 questions per language). Models answer with thinking disabled and unparsed answers count as wrong; percent correct, higher is better, random guessing 25. A small board evaluated by the GhanaNLP community.

Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2028. 3 models tracked.

Top models

#ModelScoreOverall rank
1Gemini 3 Flash (Preview) (Non-reasoning)32.65#90 (Gemini 3 Flash (Preview))
2GPT-4o26.02#357
3DeepSeek V4 Flash (Non-reasoning)26.02#155 (DeepSeek V4 Flash)

No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.

Interactive version: theaggregate.ai/benchmark?slug=nsanku-mmlu-kasem · How It Works · Data refreshed daily, snapshot 2026-11-05.