Granite 4.0 H Small: benchmark results
IBM's Apache-2.0 workhorse from the Granite 4.0 line: a 32B hybrid Mamba-2/transformer MoE with 9B active, built for low-memory enterprise agents (October 2025). Provider: IBM. Released 2025-10-01. Access: Open.
Unified ELO 1491 ± 1, rank #730 of 1392 rated models, from 301 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Bosnian | 57.76 | Average Score (%) | 96.7 |
| EuroEval Hungarian NLU - HuSST | 62.31 | Sentiment classification Score (%) | 95 |
| EuroEval Slovene NLU - MultiWikiQA SL | 70.68 | Reading comprehension Score (%) | 93.6 |
| EuroEval Finnish NLU - Scandisent FI | 92.45 | Sentiment classification Score (%) | 91.8 |
| EuroEval French NLU - ScaLA FR | 58.09 | Linguistic acceptability Score (%) | 91.8 |
| Enkrypt AI - Toxicity Risk | 0.36 | Risk Score | 91.4 |
| EuroEval English | 74.48 | Average Score (%) | 90.4 |
| EuroEval French NLU | 69.89 | NLU Average Score (%) | 90.1 |
| EuroEval French | 68.61 | Average Score (%) | 89.4 |
| EuroEval Spanish Common Sense Reasoning | 75.34 | Common Sense Reasoning Average Score (%) | 88.9 |
| EuroEval Italian Common Sense Reasoning | 70.17 | Common Sense Reasoning Average Score (%) | 88.8 |
| EuroEval French Common Sense Reasoning | 72.25 | Common Sense Reasoning Average Score (%) | 88.6 |
Interactive version: theaggregate.ai/model?slug=granite-4-0-h-small · How It Works · Data refreshed daily, snapshot 2026-09-05.