SAHARA - Context-Based QA: leaderboard
Context-based extractive question answering (squad_qa), currently Swahili only, from the SAHARA African-language LLM leaderboard (UBC-NLP, ACL 2025). The task score is the mean over the languages the task covers; Macro-F1 on a 0-100 scale, higher is better.
Metric: Macro-F1 (%). Source: huggingface.co. Saturation forecast: Around February 2027. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 84.82 | #105 |
| 2 | GPT-5.1 (2025-11-13) | 82.81 | #107 |
| 3 | Gemma 2 27B (IT) | 82.49 | #652 |
| 4 | Qwen 3.6 27B | 81.57 | #182 |
| 5 | Qwen 3.6 35B A3B | 81.36 | #238 |
| 6 | Llama 4 Scout Instruct | 81.34 | #547 |
| 7 | Qwen 3.5 27B | 81.26 | #192 |
| 8 | Gemma 2 9B (IT) | 80.91 | #795 |
| 9 | c4ai-command-a-03-2025 | 80.05 | #468 |
| 10 | Gemini 2.5 Flash | 80.03 | #255 |
| 11 | Gemma 3 27B (IT) | 79.16 | #528 |
| 12 | Gemini 2.5 Pro | 79.03 | #159 |
| 13 | GPT-5.6 Luna | 78.3 | #79 |
| 14 | Llama 3.1 8B Instruct | 78.04 | #1017 |
| 15 | Claude Sonnet 4 (20250514) | 78.02 | #223 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sahara-context-based-qa · How It Works · Data refreshed daily, snapshot 2026-11-05.