SAHARA - Cross-Lingual NLI: leaderboard
Cross-lingual natural language inference (xlni), from the SAHARA African-language LLM leaderboard (UBC-NLP, ACL 2025). The task score is the mean over the languages the task covers; Accuracy on a 0-100 scale, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 77.58 | #71 |
| 2 | Gemini 2.5 Pro | 75.48 | #159 |
| 3 | Gemini 2.5 Flash | 73.47 | #255 |
| 4 | Claude Sonnet 4 | 72.09 | #207 |
| 5 | GPT-5.6 Luna | 71.69 | #79 |
| 6 | GPT-4.1 | 69.6 | #258 |
| 7 | Claude Sonnet 4 (20250514) | 69 | #223 |
| 8 | GPT-5.1 (2025-11-13) | 68.39 | #107 |
| 9 | Qwen 3.6 27B | 66.79 | #182 |
| 10 | GPT-5 | 65.98 | #105 |
| 11 | Qwen 3.6 35B A3B | 65.79 | #238 |
| 12 | Gemma 3 27B (IT) | 65.58 | #528 |
| 13 | Qwen 3.5 27B | 64.48 | #192 |
| 14 | Gemma 3 12B (IT) | 61.09 | #672 |
| 15 | Llama 4 Scout Instruct | 59.88 | #547 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sahara-cross-lingual-nli · How It Works · Data refreshed daily, snapshot 2026-11-05.