granite-4.0-h-micro: benchmark results
IBM's Apache-2.0 3B micro-tier Granite 4.0 model (October 2025) with a hybrid Mamba-2/transformer design for low-memory, low-latency inference. Provider: IBM. Released 2025-10-02. Access: Open.
Aggregate Rating 1432 ± 11, rank #1088 of 1656 rated models, from 564 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AGC-Bench - puntuguese | 2.49 | Dataset z-score | 100 |
| AGC-Bench - futuregen | 1.51 | Dataset z-score | 97.5 |
| AGC-Bench - twistlist | 1.51 | Dataset z-score | 95.1 |
| EuroEval English Summarization - CNN Dailymail | 43.85 | chrF3++ (%) | 93.3 |
| EuroEval Spanish Summarization - Mlsum ES | 29.33 | chrF3++ (%) | 92.5 |
| Kurakurai French LLM - MBPP | 66.32 | pass@1 (%) | 90.2 |
| EuroEval Ukrainian Summarization - LR SUM UK | 31.03 | chrF3++ (%) | 89.7 |
| Kurakurai French LLM - BFCL v2 | 72.3 | Accuracy (%) | 89.7 |
| AGC-Bench - creatset | 1.37 | Dataset z-score | 89 |
| AGC-Bench - hypobench | 0.38 | Dataset z-score | 86.6 |
| Kurakurai French LLM - HumanEval+ | 65.56 | pass@1 (%) | 85.4 |
| QIMMA - ArabLegalQA | 71.68 | BERTScore F1 (%) | 84.9 |
Interactive version: theaggregate.ai/model?slug=granite-4-0-h-micro · How It Works · Data refreshed daily, snapshot 2026-10-20.