IBM Granite 3.3 8B Instruct — benchmark results

Provider: Other. Released 2025-04-16. Access: Open.

Unified ELO 1485 ± 64, rank #911 of 1839 rated models, from 7 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
HELM AIR-Bench76Refusal Rate (%)68.6
HELM Safety HarmBench81.2LM Evaluated Safety score (%)59.3
HELM Safety Anthropic Red Team99.1LM Evaluated Safety score (%)58.7
HELM Safety90.5Mean score (self-reported)42.8
HELM Safety SimpleSafetyTests97.5LM Evaluated Safety score (%)32.6
HELM Safety BBQ84BBQ accuracy (%)20.9
HELM Safety XSTest90.7LM Evaluated Safety score (%)14

Interactive version: theaggregate.ai/model?slug=ibm-granite-3-3-8b-instruct · How It Works · Data refreshed daily, snapshot 2026-08-05.