Nile-Chat-4B: benchmark results
Provider: Other. Released 2025-07-06. Access: Open.
Aggregate Rating 1416 ± 33, rank #1165 of 1656 rated models, from 36 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Granite Arabic Card - Translation - MSA→En | 37.12 | BLEU (0–100) | 100 |
| Granite Arabic Card - Translation - Dial.→En | 31.28 | BLEU (0–100) | 77.8 |
| Granite Arabic Card - Question Answering - Belebele (Dial) | 66.38 | Exact Match (%) | 66.7 |
| Granite Arabic Card - Question Answering - Global PIQA (Dial) | 72.8 | Exact Match (%) | 66.7 |
| Granite Arabic Card - Question Answering - Global PIQA (MSA) | 72 | Exact Match (%) | 66.7 |
| QIMMA - ArabCulture | 51.57 | Accuracy (%, log-likelihood multiple choice) | 64.6 |
| QIMMA - MedAraBench | 39.39 | Accuracy (%, log-likelihood multiple choice) | 63.4 |
| QIMMA - MedArabiQ MCQ | 37.86 | Accuracy (%, log-likelihood multiple choice) | 58.5 |
| Granite Arabic Card - Question Answering - DialArMMLU (Dial) | 48.77 | Exact Match (%) | 55.6 |
| Granite Arabic Card - Question Answering - INCLUDE-base (MSA) | 50.91 | Exact Match (%) | 55.6 |
| Granite Arabic Card - Translation - Average | 21.54 | BLEU (0–100) | 55.6 |
| QIMMA - GAT | 40.93 | Accuracy (%, log-likelihood multiple choice) | 50 |
Interactive version: theaggregate.ai/model?slug=nile-chat-4b · How It Works · Data refreshed daily, snapshot 2026-10-20.