Granite Arabic Card - Translation - MSA→En: leaderboard
IBM Research / MBZUAI Granite 4.0 H 3B Arabic public checkpoint card; lm-evaluation-harness five-shot evaluation. Flores-200 translation evaluated with BLEU, explicitly higher is better despite the descending-sort arrow in the average column. Source abbreviations g4.0-h-micro_hf and g3.3-2b-instruct remain source identities without an unverified checkpoint merge. BLEU (0–100); higher is better.
Metric: BLEU (0–100). Source: huggingface.co. Saturation forecast: Around May 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Nile-Chat-4B | 37.12 | #1165 |
| 2 | granite-4.0-h-3B-ar | 36.54 | #857 |
| 3 | Gemma 3 4B (IT) | 35.72 | #992 |
| 4 | Qwen 3 4B 2507 Instruct | 33.12 | #762 |
| 5 | Atlas-Chat-2B | 32.38 | #1284 |
| 6 | Falcon-H1-3B-Instruct | 32.31 | #1221 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=granite-arabic-card-translation-msa-en · How It Works · Data refreshed daily, snapshot 2026-10-20.