SAHARA - Machine Translation (African to African): leaderboard
Machine translation between African languages (mt_xx2xx), from the SAHARA African-language LLM leaderboard (UBC-NLP, ACL 2025). The task score is the mean over the languages the task covers; spBLEU-1K on a 0-100 scale, higher is better. Absolute values are low.
Metric: spBLEU-1K. Source: huggingface.co. Saturation forecast: Around October 2028. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 12.23 | #71 |
| 2 | Gemini 2.5 Pro | 8.64 | #159 |
| 3 | GPT-5.6 Luna | 6.4 | #79 |
| 4 | Gemini 2.5 Flash | 5.89 | #255 |
| 5 | Claude Sonnet 4 (20250514) | 5.15 | #223 |
| 6 | GPT-4.1 | 4.44 | #258 |
| 7 | GPT-5 | 4.25 | #105 |
| 8 | GPT-5.1 (2025-11-13) | 3.92 | #107 |
| 9 | Claude Sonnet 4 | 2.8 | #207 |
| 10 | Qwen 3.6 27B | 1.54 | #182 |
| 11 | Qwen 3.6 35B A3B | 1.4 | #238 |
| 12 | Gemma 3 27B (IT) | 1.39 | #528 |
| 13 | DeepSeek R1 Distill Llama 70B | 1.37 | #614 |
| 14 | Llama 3.1 70B Instruct | 1.28 | #567 |
| 15 | Qwen 3.5 27B | 1.07 | #192 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sahara-machine-translation-african-to-african · How It Works · Data refreshed daily, snapshot 2026-11-05.