Granite 4.2 3B: benchmark results
IBM's Granite 4.2 3B, the compact reasoning model in the Granite 4.2 family, with native chain-of-thought and switchable full, non-thinking and low-effort thinking modes. Provider: IBM. Released 2026-08-25. Access: Open.
Unified ELO 1501 ± 34, rank #773 of 1639 rated models, from 84 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA-Omniscience Hallucination Rate | 26.34 | Hallucination Rate (%) | 94.9 |
| AA-Omniscience Index - Law | -7.4 | Omniscience Index | 73.9 |
| AA-Omniscience Index - Humanities & Social Sciences | -7.2 | Omniscience Index | 71.2 |
| AA-Omniscience Index - Business | -9.1 | Omniscience Index | 70.9 |
| AA-Omniscience Index - Health | -19.9 | Omniscience Index | 64.6 |
| AA-Omniscience Index - Software Engineering (SWE) - Julia | -24 | Omniscience Index | 64.4 |
| AA Omniscience | -14.72 | Score | 62.3 |
| AA-Omniscience Index - Software Engineering (SWE) - Java | -33 | Omniscience Index | 61.3 |
| AA-Omniscience Index - Science, Engineering & Mathematics | -13.7 | Omniscience Index | 58.3 |
| AA-Omniscience Index - Software Engineering (SWE) - Go | -22 | Omniscience Index | 57.9 |
| AA-Omniscience Index - Software Engineering (SWE) - Kotlin | -24 | Omniscience Index | 57.3 |
| AA-Omniscience Index - Software Engineering (SWE) - Python | -21.5 | Omniscience Index | 57.3 |
Interactive version: theaggregate.ai/model?slug=granite-4-2-3b · How It Works · Data refreshed daily, snapshot 2026-10-09.