granite-3.0-2B-instruct: benchmark results
IBM's Granite 3.0 2B Instruct, fine-tuned from Granite-3.0-2B-Base using supervised fine-tuning, reinforcement learning and model merging, supporting 12 languages. Provider: IBM. Released 2024-10-21. Access: Open.
Aggregate Rating 1304 ± 20, rank #1471 of 1656 rated models, from 94 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Enkrypt AI - Toxicity Risk | 0.5 | Risk Score | 90 |
| Enkrypt AI - Safety Risk | 22.78 | Risk Score | 70.7 |
| Open LLM Leaderboard - IFEval | 51.4 | Score | 61.3 |
| EuroEval Portuguese NLU - HAREM | 43.78 | Micro F1 excluding MISC (%) | 59 |
| Open PL LLM - EQ-Bench | 33.06 | EQ-Bench Score | 58.4 |
| Open LLM Leaderboard - GPQA | 29.95 | Score | 55.6 |
| Enkrypt AI - Jailbreak Risk | 8.75 | Risk Score | 53.4 |
| DeEscalWild | 32.4 | Realism score (0-100) from a Gemini 3.1 Pro judge rating beh | 50 |
| DeEscalWild - METEOR | 13.5 | METEOR (0-100) of the generated civilian turns against the r | 50 |
| DeEscalWild - ROUGE-L | 10.2 | ROUGE-L (0-100) of the generated civilian turns against the | 50 |
| DeEscalWild - Realism (GPT-5.4 Judge) | 31.8 | Realism score (0-100) from a GPT-5.4 judge rating behavioura | 50 |
| EuroEval Dutch NLU - DBRD | 86.12 | Matthews correlation coefficient (%) | 47.8 |
Interactive version: theaggregate.ai/model?slug=granite-3-0-2b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.