SAHARA - MCCR Cluster: leaderboard
Multiple-choice, comprehension and reasoning cluster score from the SAHARA African-language LLM leaderboard (UBC-NLP, ACL 2025): mean of general-knowledge accuracy, math word-problem exact match, Belebele reading-comprehension accuracy and context-based QA macro-F1. Scale 0-100, higher is better.
Metric: Cluster score (mean of task scores). Source: huggingface.co. Saturation forecast: Around February 2027. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 78.58 | #71 |
| 2 | Gemini 2.5 Pro | 77.56 | #159 |
| 3 | GPT-5 | 75.77 | #105 |
| 4 | GPT-5.1 (2025-11-13) | 73.64 | #107 |
| 5 | Gemini 2.5 Flash | 72.41 | #255 |
| 6 | GPT-5.6 Luna | 68.5 | #79 |
| 7 | Claude Sonnet 4 (20250514) | 60.53 | #223 |
| 8 | Qwen 3.5 27B | 54.61 | #192 |
| 9 | Qwen 3.6 27B | 54.06 | #182 |
| 10 | Claude Sonnet 4 | 52.41 | #207 |
| 11 | Qwen 3.6 35B A3B | 51.55 | #238 |
| 12 | GPT-4.1 | 50.98 | #258 |
| 13 | Llama 4 Scout Instruct | 47.36 | #547 |
| 14 | c4ai-command-a-03-2025 | 45.55 | #468 |
| 15 | Llama 3.3 70B Instruct | 44.77 | #541 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sahara-mccr-cluster · How It Works · Data refreshed daily, snapshot 2026-11-05.