gemma-2B (IT): benchmark results
Google's Gemma 2B instruction-tuned model, released under the Gemma license in February 2024; the linked model card provides no further architecture or training details. Provider: Google. Released 2024-02-21. Access: Open.
Aggregate Rating 1235 ± 12, rank #1559 of 1656 rated models, from 193 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SALAD-Bench | 73.12 | Average Safety Score (%) | 84.8 |
| SALAD-Bench Attack | 50.34 | Safety Score (%) | 84.8 |
| LLM Trustworthy - Privacy | 88.77 | Trust Score (%) | 79.2 |
| SALAD-Bench Base | 95.9 | Safety Score (%) | 78.8 |
| LLM Trustworthy - Toxicity | 77.07 | Trust Score (%) | 70.8 |
| LLM Trustworthy - Ethics | 75.03 | Trust Score (%) | 62.5 |
| AI-Secure LLM Trustworthy Leaderboard | 0.67 | Trustworthy average | 58.3 |
| LLM Trustworthy Leaderboard | 67.18 | Average Trust Score (%) | 58.3 |
| LLM Trustworthy - Fairness | 93.02 | Trust Score (%) | 54.2 |
| Open CoT - LogiQA | 4.31 | CoT Gain (%) | 54.2 |
| Open CoT - LSAT Analytical Reasoning | 3.91 | CoT Gain (%) | 50.8 |
| Open Japanese LLM - Xlsum JA Rouge1 | 25.26 | Score (%) | 46.9 |
Interactive version: theaggregate.ai/model?slug=gemma-2b-it · How It Works · Data refreshed daily, snapshot 2026-10-20.