h2o-danube3-4B-chat: benchmark results
H2O.ai's h2o-danube3-4b-chat, a 4B chat model using an adjusted Llama 2 architecture with the Mistral tokenizer, trained for an 8,192-token context and designed to run offline on phones. Provider: H2O.ai. Released 2024-07-04. Access: Open.
Aggregate Rating 1303 ± 58, rank #1475 of 1656 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Enkrypt AI - Jailbreak Risk | 6.71 | Risk Score | 68.2 |
| Open Portuguese LLM - OAB Exams | 38.13 | Accuracy (%) | 37.3 |
| Open Portuguese LLM - ENEM | 52.55 | Accuracy (%) | 36.1 |
| Open LLM Leaderboard - IFEval | 36.29 | Score | 34.7 |
| Open Portuguese LLM - FaQuAD NLI | 51.19 | Macro F1 (%) | 32.7 |
| Open Portuguese LLM - BLUEX | 38.25 | Accuracy (%) | 31.5 |
| Open LLM Leaderboard - MuSR | 37.81 | Score | 28.5 |
| Open Portuguese LLM - ASSIN2 RTE | 76.92 | Macro F1 (%) | 28.4 |
| Enkrypt AI - Toxicity Risk | 5.91 | Risk Score | 26.9 |
| Enkrypt AI - Safety Risk | 35.47 | Risk Score | 23.8 |
| Open LLM Leaderboard - MATH Level 5 | 4.08 | Score | 23.5 |
| Open LLM Leaderboard - MMLU-Pro | 22.28 | Score | 23.1 |
Interactive version: theaggregate.ai/model?slug=h2o-danube3-4b-chat · How It Works · Data refreshed daily, snapshot 2026-10-20.