Granite 4.0 Micro: benchmark results
IBM's Apache-2.0 3B dense member of the Granite 4.0 family (October 2025), the conventional-transformer option for edge and local deployment. Provider: IBM. Released 2025-10-02. Access: Open.
Aggregate Rating 1472 ± 22, rank #882 of 1656 rated models, from 694 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval English Summarization - CNN Dailymail | 45.11 | chrF3++ (%) | 98.7 |
| EuroEval Ukrainian Summarization - LR SUM UK | 31.58 | chrF3++ (%) | 97.9 |
| HELM AIR-Bench 2024 - #10.12: Bodily destruction | 96.67 | Refusal Rate (%) | 94.8 |
| HELM AIR-Bench 2024 - #14.25: Personality | 100 | Refusal Rate (%) | 92.4 |
| HELM AIR-Bench 2024 - #14.28: Pregnancy status | 100 | Refusal Rate (%) | 91.9 |
| HELM AIR-Bench 2024 - #10.13: Bodily mutilation | 90 | Refusal Rate (%) | 91.3 |
| HELM AIR-Bench 2024 - #17.2: Suggestive | 73.33 | Refusal Rate (%) | 91.3 |
| HELM AIR-Bench 2024 - #3.11: System/Website impairment | 100 | Refusal Rate (%) | 91.3 |
| EuroEval Italian NLU - ScaLA IT | 45.89 | Matthews correlation coefficient (%) | 90.5 |
| HELM AIR-Bench 2024 - #10: Depicting Violence | 72.67 | Refusal Rate (%) | 90.1 |
| EuroEval French NLU - ScaLA FR | 55.87 | Matthews correlation coefficient (%) | 89.4 |
| Enkrypt AI - Toxicity Risk | 0.55 | Risk Score | 89.1 |
Interactive version: theaggregate.ai/model?slug=granite-4-0-micro · How It Works · Data refreshed daily, snapshot 2026-10-20.